Initial release: technical documentation, quantization scripts, and NPU benchmarks (part 2)
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- .gitattributes +1 -0
- results/MobileNetV4/machine_2/mobilenetv4_machine_2_stability_100_runs.png +0 -0
- results/MobileNetV5/machine_1/benchmark_machine_1_v5_standard.csv +0 -0
- results/MobileNetV5/machine_1/benchmark_machine_1_v5_zero_fp32.csv +0 -0
- results/MobileNetV5/machine_1/latency_comparison_machine_1.png +3 -0
- results/MobileNetV5/machine_1/machine_1_scientific_analysis.csv +326 -0
- results/MobileNetV5/machine_2/benchmark_machine_2_v5.csv +0 -0
- results/MobileNetV5/machine_2/benchmark_machine_2_v5_100_runs.csv +601 -0
- results/MobileNetV5/machine_2/benchmark_machine_2_v5_comparative.csv +601 -0
- results/MobileNetV5/machine_2/benchmark_machine_2_v5_native_100_runs.csv +201 -0
- results/MobileNetV5/machine_2/mobilenetv5_machine_2_executive_dashboard.png +0 -0
- results/MobileNetV5/machine_2/mobilenetv5_machine_2_hardware_latency_comparison.png +0 -0
- results/MobileNetV5/machine_2/mobilenetv5_machine_2_latency_comparison.png +0 -0
- results/MobileNetV5/machine_2/mobilenetv5_machine_2_native_stability.png +0 -0
- results/MobileNetV5/machine_2/mobilenetv5_machine_2_npu_stability_distribution.png +0 -0
- results/MobileNetV5/machine_2/mobilenetv5_machine_2_pareto_accuracy_vs_latency.png +0 -0
- results/MobileNetV5/machine_2/mobilenetv5_machine_2_quantization_paradigm_comparison.png +0 -0
- results/MobileNetV5/machine_2/mobilenetv5_machine_2_stability_100_runs.png +0 -0
- results/ScaleDetection/machine_1/benchmark_machine_1_scale.csv +401 -0
- results/ScaleDetection/machine_1/scaledetection_machine_1_comparison.png +0 -0
- results/ScaleDetection/machine_2/benchmark_machine_2_scale_comparative.csv +601 -0
- results/ScaleDetection/machine_2/benchmark_machine_2_scale_native_100_runs.csv +201 -0
- results/ScaleDetection/machine_2/scale_metrology_detection_profiling.png +0 -0
- results/ScaleDetection/machine_2/scaledetection_machine_2_native_stability.png +0 -0
- scripts/benchmark_npu_onnx.py +189 -0
- scripts/benchmark_npu_qairt.py +168 -0
- scripts/benchmark_scale_machine_1.py +187 -0
- scripts/benchmark_snapdragon_native_binaries.py +94 -0
- scripts/build_all_qairt_models.py +58 -0
- scripts/compile_intel_aot.py +164 -0
- scripts/compile_mediatek_aot.py +178 -0
- scripts/compile_qairt_aot.py +156 -0
- scripts/compile_tensor_aot.py +167 -0
- scripts/generate_benchmark_plots_v5.R +629 -0
- scripts/generate_scientific_graphs.py +135 -0
- scripts/generate_snapdragon_native_plots.R +153 -0
- scripts/generate_snapdragon_plots_v4.R +187 -0
- scripts/generate_workstation_plots_scale.R +62 -0
- scripts/generate_workstation_plots_v4.R +66 -0
- scripts/litho_reference_engine.py +100 -0
- scripts/onnx_v4l_rocks_int4.py +101 -0
- scripts/onnx_v4l_rocks_int8.py +100 -0
- scripts/onnx_v4s_scale_int4.py +71 -0
- scripts/onnx_v4s_scale_int8.py +66 -0
- scripts/onnx_v5_rocks_int4.py +71 -0
- scripts/onnx_v5_rocks_int8.py +67 -0
- scripts/prepare_calibration.py +50 -0
- scripts/tflite_v4l_rocks_int4.py +98 -0
- scripts/tflite_v4l_rocks_int8.py +59 -0
- scripts/tflite_v4s_scale_int4.py +46 -0
.gitattributes
CHANGED
|
@@ -243,3 +243,4 @@ official_val_208/raw_256/sample_0205_c102.raw filter=lfs diff=lfs merge=lfs -tex
|
|
| 243 |
official_val_208/raw_256/sample_0206_c103.raw filter=lfs diff=lfs merge=lfs -text
|
| 244 |
official_val_208/raw_256/sample_0207_c103.raw filter=lfs diff=lfs merge=lfs -text
|
| 245 |
results/consolidated/images/rock_detection_demo.gif filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
| 243 |
official_val_208/raw_256/sample_0206_c103.raw filter=lfs diff=lfs merge=lfs -text
|
| 244 |
official_val_208/raw_256/sample_0207_c103.raw filter=lfs diff=lfs merge=lfs -text
|
| 245 |
results/consolidated/images/rock_detection_demo.gif filter=lfs diff=lfs merge=lfs -text
|
| 246 |
+
results/MobileNetV5/machine_1/latency_comparison_machine_1.png filter=lfs diff=lfs merge=lfs -text
|
results/MobileNetV4/machine_2/mobilenetv4_machine_2_stability_100_runs.png
ADDED
|
results/MobileNetV5/machine_1/benchmark_machine_1_v5_standard.csv
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
results/MobileNetV5/machine_1/benchmark_machine_1_v5_zero_fp32.csv
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
results/MobileNetV5/machine_1/latency_comparison_machine_1.png
ADDED
|
Git LFS Details
|
results/MobileNetV5/machine_1/machine_1_scientific_analysis.csv
ADDED
|
@@ -0,0 +1,326 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Model,Task,Format,Backend,Inference_ID,Latency_ms,Accuracy
|
| 2 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,1,3455.014,80.33%
|
| 3 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,2,3568.611,80.33%
|
| 4 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,3,3388.623,80.33%
|
| 5 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,4,3312.461,80.33%
|
| 6 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,5,3326.35,80.33%
|
| 7 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,6,3311.896,80.33%
|
| 8 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,7,3299.43,80.33%
|
| 9 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,8,3297.852,80.33%
|
| 10 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,9,3266.191,80.33%
|
| 11 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,10,3307.54,80.33%
|
| 12 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,11,3564.361,80.33%
|
| 13 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,12,3623.355,80.33%
|
| 14 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,13,3342.59,80.33%
|
| 15 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,14,3463.733,80.33%
|
| 16 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,15,3366.917,80.33%
|
| 17 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,16,3282.197,80.33%
|
| 18 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,17,3253.491,80.33%
|
| 19 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,18,3254.939,80.33%
|
| 20 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,19,3271.902,80.33%
|
| 21 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,20,3293.22,80.33%
|
| 22 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,21,3373.727,80.33%
|
| 23 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,22,3332.897,80.33%
|
| 24 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,23,3558.709,80.33%
|
| 25 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,24,3316.841,80.33%
|
| 26 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,25,3368.575,80.33%
|
| 27 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,26,3307.701,80.33%
|
| 28 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,27,3349.411,80.33%
|
| 29 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,28,3372.588,80.33%
|
| 30 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,29,3242.174,80.33%
|
| 31 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,30,3271.37,80.33%
|
| 32 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,31,3614.175,80.33%
|
| 33 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,32,3364.028,80.33%
|
| 34 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,33,3327.917,80.33%
|
| 35 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,34,3439.419,80.33%
|
| 36 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,35,3261.198,80.33%
|
| 37 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,36,3312.623,80.33%
|
| 38 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,37,3268.368,80.33%
|
| 39 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,38,3294.491,80.33%
|
| 40 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,39,3241.71,80.33%
|
| 41 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,40,3260.713,80.33%
|
| 42 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,41,3402.906,80.33%
|
| 43 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,42,3250.862,80.33%
|
| 44 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,43,3316.837,80.33%
|
| 45 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,44,3283.949,80.33%
|
| 46 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,45,3286.371,80.33%
|
| 47 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,46,3259.829,80.33%
|
| 48 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,47,3265.471,80.33%
|
| 49 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,48,3257.465,80.33%
|
| 50 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,49,3318.434,80.33%
|
| 51 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,50,3348.687,80.33%
|
| 52 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,51,3400.488,80.33%
|
| 53 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,52,3282.33,80.33%
|
| 54 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,53,3359.068,80.33%
|
| 55 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,54,3242.714,80.33%
|
| 56 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,55,3266.401,80.33%
|
| 57 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,56,3317.704,80.33%
|
| 58 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,57,3284.812,80.33%
|
| 59 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,58,3255.415,80.33%
|
| 60 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,59,3291.121,80.33%
|
| 61 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,60,3382.509,80.33%
|
| 62 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,61,3277.985,80.33%
|
| 63 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,62,3300.814,80.33%
|
| 64 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,63,3221.879,80.33%
|
| 65 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,64,3251.08,80.33%
|
| 66 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,65,3314.094,80.33%
|
| 67 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,66,3254.096,80.33%
|
| 68 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,67,3292.147,80.33%
|
| 69 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,68,3304.533,80.33%
|
| 70 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,69,3337.636,80.33%
|
| 71 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,70,3368.797,80.33%
|
| 72 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,71,3269.691,80.33%
|
| 73 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,72,3301.664,80.33%
|
| 74 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,73,3417.699,80.33%
|
| 75 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,74,3290.676,80.33%
|
| 76 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,75,3253.862,80.33%
|
| 77 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,76,3277.07,80.33%
|
| 78 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,77,3329.123,80.33%
|
| 79 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,78,3292.42,80.33%
|
| 80 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,79,3280.966,80.33%
|
| 81 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,80,3411.981,80.33%
|
| 82 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,81,3262.166,80.33%
|
| 83 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,82,3289.289,80.33%
|
| 84 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,83,3296.536,80.33%
|
| 85 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,84,3244.035,80.33%
|
| 86 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,85,3273.154,80.33%
|
| 87 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,86,3321.101,80.33%
|
| 88 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,87,3303.358,80.33%
|
| 89 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,88,3268.613,80.33%
|
| 90 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,89,3326.172,80.33%
|
| 91 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,90,3361.535,80.33%
|
| 92 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,91,3307.581,80.33%
|
| 93 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,92,3311.221,80.33%
|
| 94 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,93,3265.289,80.33%
|
| 95 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,94,3267.092,80.33%
|
| 96 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,95,3278.452,80.33%
|
| 97 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,96,3285.282,80.33%
|
| 98 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,97,3270.283,80.33%
|
| 99 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,98,3290.639,80.33%
|
| 100 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,99,3327.386,80.33%
|
| 101 |
+
MobileNetV5,Rock_Classification,W4A8,Tensor_Cores_INT4,100,3316.209,80.33%
|
| 102 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,1,147.222,83.33%
|
| 103 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,2,136.045,83.33%
|
| 104 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,3,151.507,83.33%
|
| 105 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,4,138.462,83.33%
|
| 106 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,5,138.778,83.33%
|
| 107 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,6,136.91,83.33%
|
| 108 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,7,146.682,83.33%
|
| 109 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,8,136.47,83.33%
|
| 110 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,9,141.745,83.33%
|
| 111 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,10,173.79,83.33%
|
| 112 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,11,136.645,83.33%
|
| 113 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,12,141.739,83.33%
|
| 114 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,13,145.448,83.33%
|
| 115 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,14,138.025,83.33%
|
| 116 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,15,140.214,83.33%
|
| 117 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,16,145.246,83.33%
|
| 118 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,17,141.727,83.33%
|
| 119 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,18,149.556,83.33%
|
| 120 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,19,138.33,83.33%
|
| 121 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,20,144.253,83.33%
|
| 122 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,21,142.304,83.33%
|
| 123 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,22,130.939,83.33%
|
| 124 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,23,139.774,83.33%
|
| 125 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,24,141.757,83.33%
|
| 126 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,25,134.124,83.33%
|
| 127 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,26,144.765,83.33%
|
| 128 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,27,133.154,83.33%
|
| 129 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,28,138.971,83.33%
|
| 130 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,29,137.849,83.33%
|
| 131 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,30,163.003,83.33%
|
| 132 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,31,133.727,83.33%
|
| 133 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,32,139.466,83.33%
|
| 134 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,33,136.838,83.33%
|
| 135 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,34,132.941,83.33%
|
| 136 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,35,133.634,83.33%
|
| 137 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,36,138.988,83.33%
|
| 138 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,37,150.168,83.33%
|
| 139 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,38,131.96,83.33%
|
| 140 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,39,136.762,83.33%
|
| 141 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,40,138.537,83.33%
|
| 142 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,41,134.014,83.33%
|
| 143 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,42,137.126,83.33%
|
| 144 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,43,139.387,83.33%
|
| 145 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,44,134.45,83.33%
|
| 146 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,45,133.252,83.33%
|
| 147 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,46,136.438,83.33%
|
| 148 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,47,134.517,83.33%
|
| 149 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,48,139.602,83.33%
|
| 150 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,49,163.613,83.33%
|
| 151 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,50,146.091,83.33%
|
| 152 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,51,132.786,83.33%
|
| 153 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,52,138.509,83.33%
|
| 154 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,53,140.953,83.33%
|
| 155 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,54,131.775,83.33%
|
| 156 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,55,135.209,83.33%
|
| 157 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,56,142.771,83.33%
|
| 158 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,57,135.556,83.33%
|
| 159 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,58,129.606,83.33%
|
| 160 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,59,136.18,83.33%
|
| 161 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,60,140.993,83.33%
|
| 162 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,61,149.847,83.33%
|
| 163 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,62,137.443,83.33%
|
| 164 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,63,136.662,83.33%
|
| 165 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,64,137.421,83.33%
|
| 166 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,65,134.453,83.33%
|
| 167 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,66,145.797,83.33%
|
| 168 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,67,153.944,83.33%
|
| 169 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,68,148.057,83.33%
|
| 170 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,69,168.34,83.33%
|
| 171 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,70,138.992,83.33%
|
| 172 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,71,137.341,83.33%
|
| 173 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,72,135.25,83.33%
|
| 174 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,73,136.118,83.33%
|
| 175 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,74,137.048,83.33%
|
| 176 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,75,137.138,83.33%
|
| 177 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,76,137.519,83.33%
|
| 178 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,77,136.011,83.33%
|
| 179 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,78,139.331,83.33%
|
| 180 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,79,133.262,83.33%
|
| 181 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,80,133.835,83.33%
|
| 182 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,81,146.89,83.33%
|
| 183 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,82,134.76,83.33%
|
| 184 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,83,136.22,83.33%
|
| 185 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,84,136.492,83.33%
|
| 186 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,85,131.898,83.33%
|
| 187 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,86,137.142,83.33%
|
| 188 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,87,136.835,83.33%
|
| 189 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,88,141.544,83.33%
|
| 190 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,89,168.301,83.33%
|
| 191 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,90,138.502,83.33%
|
| 192 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,91,143.783,83.33%
|
| 193 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,92,132.115,83.33%
|
| 194 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,93,131.854,83.33%
|
| 195 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,94,138.378,83.33%
|
| 196 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,95,142.207,83.33%
|
| 197 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,96,139.874,83.33%
|
| 198 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,97,149.606,83.33%
|
| 199 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,98,132.178,83.33%
|
| 200 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,99,135.183,83.33%
|
| 201 |
+
MobileNetV5,Rock_Classification,INT8,Tensor_Cores_INT8,100,145.268,83.33%
|
| 202 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,1,3363.123,80.33%
|
| 203 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,2,3452.08,80.33%
|
| 204 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,3,3362.233,80.33%
|
| 205 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,4,3358.338,80.33%
|
| 206 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,5,3325.73,80.33%
|
| 207 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,6,3353.662,80.33%
|
| 208 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,7,3364.378,80.33%
|
| 209 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,8,3352.634,80.33%
|
| 210 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,9,3325.656,80.33%
|
| 211 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,10,3379.99,80.33%
|
| 212 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,11,3382.554,80.33%
|
| 213 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,12,3536.887,80.33%
|
| 214 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,13,3324.387,80.33%
|
| 215 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,14,3300.808,80.33%
|
| 216 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,15,3400.228,80.33%
|
| 217 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,16,3344.73,80.33%
|
| 218 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,17,3348.927,80.33%
|
| 219 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,18,3329.126,80.33%
|
| 220 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,19,3334.868,80.33%
|
| 221 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,20,3322.692,80.33%
|
| 222 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,21,3486.679,80.33%
|
| 223 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,22,3400.648,80.33%
|
| 224 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,23,3357.35,80.33%
|
| 225 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,24,3386.929,80.33%
|
| 226 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,25,3356.549,80.33%
|
| 227 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,26,3317.325,80.33%
|
| 228 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,27,3394.19,80.33%
|
| 229 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,28,3343.68,80.33%
|
| 230 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,29,3339.526,80.33%
|
| 231 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,30,3426.658,80.33%
|
| 232 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,31,3534.172,80.33%
|
| 233 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,32,3353.723,80.33%
|
| 234 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,33,3433.833,80.33%
|
| 235 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,34,3339.33,80.33%
|
| 236 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,35,3366.505,80.33%
|
| 237 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,36,3371.125,80.33%
|
| 238 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,37,3404.507,80.33%
|
| 239 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,38,3324.451,80.33%
|
| 240 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,39,3373.493,80.33%
|
| 241 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,40,3391.213,80.33%
|
| 242 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,41,3327.308,80.33%
|
| 243 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,42,3460.567,80.33%
|
| 244 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,43,3318.048,80.33%
|
| 245 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,44,3373.088,80.33%
|
| 246 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,45,3444.845,80.33%
|
| 247 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,46,3327.002,80.33%
|
| 248 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,47,3345.894,80.33%
|
| 249 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,48,3402.777,80.33%
|
| 250 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,49,3323.291,80.33%
|
| 251 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,50,3609.208,80.33%
|
| 252 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,51,3341.108,80.33%
|
| 253 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,52,3373.795,80.33%
|
| 254 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,53,3353.51,80.33%
|
| 255 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,54,3423.651,80.33%
|
| 256 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,55,3327.27,80.33%
|
| 257 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,56,3336.924,80.33%
|
| 258 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,57,3389.693,80.33%
|
| 259 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,58,3344.71,80.33%
|
| 260 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,59,3309.359,80.33%
|
| 261 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,60,3531.595,80.33%
|
| 262 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,61,3321.095,80.33%
|
| 263 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,62,3365.415,80.33%
|
| 264 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,63,3363.55,80.33%
|
| 265 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,64,3324.748,80.33%
|
| 266 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,65,3355.942,80.33%
|
| 267 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,66,3375.211,80.33%
|
| 268 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,67,3380.331,80.33%
|
| 269 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,68,3344.698,80.33%
|
| 270 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,69,3447.858,80.33%
|
| 271 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,70,3348.039,80.33%
|
| 272 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,71,3357.308,80.33%
|
| 273 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,72,3331.334,80.33%
|
| 274 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,73,3383.089,80.33%
|
| 275 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,74,3387.214,80.33%
|
| 276 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,75,3515.999,80.33%
|
| 277 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,76,3374.874,80.33%
|
| 278 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,77,3371.105,80.33%
|
| 279 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,78,3356.972,80.33%
|
| 280 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,79,3508.61,80.33%
|
| 281 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,80,3345.502,80.33%
|
| 282 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,81,3352.796,80.33%
|
| 283 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,82,3322.594,80.33%
|
| 284 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,83,3303.418,80.33%
|
| 285 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,84,3344.972,80.33%
|
| 286 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,85,3394.967,80.33%
|
| 287 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,86,3411.335,80.33%
|
| 288 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,87,3375.633,80.33%
|
| 289 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,88,3477.235,80.33%
|
| 290 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,89,3344.844,80.33%
|
| 291 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,90,3345.127,80.33%
|
| 292 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,91,3365.361,80.33%
|
| 293 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,92,3320.226,80.33%
|
| 294 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,93,3333.748,80.33%
|
| 295 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,94,3376.776,80.33%
|
| 296 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,95,3364.157,80.33%
|
| 297 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,96,3355.369,80.33%
|
| 298 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,97,3314.774,80.33%
|
| 299 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,98,3499.875,80.33%
|
| 300 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,99,3364.285,80.33%
|
| 301 |
+
MobileNetV5,Rock_Classification,W4A8,CPU_x64,100,3377.339,80.33%
|
| 302 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,1,3021.785,83.33%
|
| 303 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,2,2986.321,83.33%
|
| 304 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,3,3053.214,83.33%
|
| 305 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,4,3051.105,83.33%
|
| 306 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,5,2991.957,83.33%
|
| 307 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,6,2981.857,83.33%
|
| 308 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,7,3021.893,83.33%
|
| 309 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,8,3194.612,83.33%
|
| 310 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,9,3028.088,83.33%
|
| 311 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,10,3046.525,83.33%
|
| 312 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,11,3002.352,83.33%
|
| 313 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,12,2951.494,83.33%
|
| 314 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,13,2985.389,83.33%
|
| 315 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,14,2973.823,83.33%
|
| 316 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,15,2976.263,83.33%
|
| 317 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,16,2966.912,83.33%
|
| 318 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,17,2991.649,83.33%
|
| 319 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,18,3016.905,83.33%
|
| 320 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,19,3109.938,83.33%
|
| 321 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,20,2990.317,83.33%
|
| 322 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,21,3026.759,83.33%
|
| 323 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,22,2953.867,83.33%
|
| 324 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,23,3019.777,83.33%
|
| 325 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,24,3075.746,83.33%
|
| 326 |
+
MobileNetV5,Rock_Classification,INT8,CPU_x64,25,3074.029,83.33%
|
results/MobileNetV5/machine_2/benchmark_machine_2_v5.csv
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
results/MobileNetV5/machine_2/benchmark_machine_2_v5_100_runs.csv
ADDED
|
@@ -0,0 +1,601 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Model,Task,Format,Backend,Inference_ID,Latency_ms,Accuracy
|
| 2 |
+
MobileNetV5,Rock_Classification,INT8,NPU,1,441.122,83.33%
|
| 3 |
+
MobileNetV5,Rock_Classification,INT8,NPU,2,439.597,83.33%
|
| 4 |
+
MobileNetV5,Rock_Classification,INT8,NPU,3,455.340,83.33%
|
| 5 |
+
MobileNetV5,Rock_Classification,INT8,NPU,4,457.711,83.33%
|
| 6 |
+
MobileNetV5,Rock_Classification,INT8,NPU,5,462.187,83.33%
|
| 7 |
+
MobileNetV5,Rock_Classification,INT8,NPU,6,438.766,83.33%
|
| 8 |
+
MobileNetV5,Rock_Classification,INT8,NPU,7,444.756,83.33%
|
| 9 |
+
MobileNetV5,Rock_Classification,INT8,NPU,8,442.980,83.33%
|
| 10 |
+
MobileNetV5,Rock_Classification,INT8,NPU,9,454.924,83.33%
|
| 11 |
+
MobileNetV5,Rock_Classification,INT8,NPU,10,444.807,83.33%
|
| 12 |
+
MobileNetV5,Rock_Classification,INT8,NPU,11,439.322,83.33%
|
| 13 |
+
MobileNetV5,Rock_Classification,INT8,NPU,12,437.309,83.33%
|
| 14 |
+
MobileNetV5,Rock_Classification,INT8,NPU,13,428.695,83.33%
|
| 15 |
+
MobileNetV5,Rock_Classification,INT8,NPU,14,484.089,83.33%
|
| 16 |
+
MobileNetV5,Rock_Classification,INT8,NPU,15,441.683,83.33%
|
| 17 |
+
MobileNetV5,Rock_Classification,INT8,NPU,16,435.857,83.33%
|
| 18 |
+
MobileNetV5,Rock_Classification,INT8,NPU,17,438.977,83.33%
|
| 19 |
+
MobileNetV5,Rock_Classification,INT8,NPU,18,447.855,83.33%
|
| 20 |
+
MobileNetV5,Rock_Classification,INT8,NPU,19,441.495,83.33%
|
| 21 |
+
MobileNetV5,Rock_Classification,INT8,NPU,20,436.225,83.33%
|
| 22 |
+
MobileNetV5,Rock_Classification,INT8,NPU,21,454.622,83.33%
|
| 23 |
+
MobileNetV5,Rock_Classification,INT8,NPU,22,437.617,83.33%
|
| 24 |
+
MobileNetV5,Rock_Classification,INT8,NPU,23,446.211,83.33%
|
| 25 |
+
MobileNetV5,Rock_Classification,INT8,NPU,24,459.077,83.33%
|
| 26 |
+
MobileNetV5,Rock_Classification,INT8,NPU,25,482.489,83.33%
|
| 27 |
+
MobileNetV5,Rock_Classification,INT8,NPU,26,445.245,83.33%
|
| 28 |
+
MobileNetV5,Rock_Classification,INT8,NPU,27,504.224,83.33%
|
| 29 |
+
MobileNetV5,Rock_Classification,INT8,NPU,28,450.616,83.33%
|
| 30 |
+
MobileNetV5,Rock_Classification,INT8,NPU,29,432.656,83.33%
|
| 31 |
+
MobileNetV5,Rock_Classification,INT8,NPU,30,453.460,83.33%
|
| 32 |
+
MobileNetV5,Rock_Classification,INT8,NPU,31,433.928,83.33%
|
| 33 |
+
MobileNetV5,Rock_Classification,INT8,NPU,32,442.961,83.33%
|
| 34 |
+
MobileNetV5,Rock_Classification,INT8,NPU,33,441.610,83.33%
|
| 35 |
+
MobileNetV5,Rock_Classification,INT8,NPU,34,455.567,83.33%
|
| 36 |
+
MobileNetV5,Rock_Classification,INT8,NPU,35,441.718,83.33%
|
| 37 |
+
MobileNetV5,Rock_Classification,INT8,NPU,36,439.802,83.33%
|
| 38 |
+
MobileNetV5,Rock_Classification,INT8,NPU,37,446.790,83.33%
|
| 39 |
+
MobileNetV5,Rock_Classification,INT8,NPU,38,440.145,83.33%
|
| 40 |
+
MobileNetV5,Rock_Classification,INT8,NPU,39,447.802,83.33%
|
| 41 |
+
MobileNetV5,Rock_Classification,INT8,NPU,40,469.870,83.33%
|
| 42 |
+
MobileNetV5,Rock_Classification,INT8,NPU,41,456.169,83.33%
|
| 43 |
+
MobileNetV5,Rock_Classification,INT8,NPU,42,448.030,83.33%
|
| 44 |
+
MobileNetV5,Rock_Classification,INT8,NPU,43,435.400,83.33%
|
| 45 |
+
MobileNetV5,Rock_Classification,INT8,NPU,44,437.904,83.33%
|
| 46 |
+
MobileNetV5,Rock_Classification,INT8,NPU,45,450.468,83.33%
|
| 47 |
+
MobileNetV5,Rock_Classification,INT8,NPU,46,442.219,83.33%
|
| 48 |
+
MobileNetV5,Rock_Classification,INT8,NPU,47,502.808,83.33%
|
| 49 |
+
MobileNetV5,Rock_Classification,INT8,NPU,48,439.923,83.33%
|
| 50 |
+
MobileNetV5,Rock_Classification,INT8,NPU,49,450.743,83.33%
|
| 51 |
+
MobileNetV5,Rock_Classification,INT8,NPU,50,452.326,83.33%
|
| 52 |
+
MobileNetV5,Rock_Classification,INT8,NPU,51,445.223,83.33%
|
| 53 |
+
MobileNetV5,Rock_Classification,INT8,NPU,52,444.033,83.33%
|
| 54 |
+
MobileNetV5,Rock_Classification,INT8,NPU,53,433.950,83.33%
|
| 55 |
+
MobileNetV5,Rock_Classification,INT8,NPU,54,442.803,83.33%
|
| 56 |
+
MobileNetV5,Rock_Classification,INT8,NPU,55,441.160,83.33%
|
| 57 |
+
MobileNetV5,Rock_Classification,INT8,NPU,56,444.141,83.33%
|
| 58 |
+
MobileNetV5,Rock_Classification,INT8,NPU,57,452.582,83.33%
|
| 59 |
+
MobileNetV5,Rock_Classification,INT8,NPU,58,454.525,83.33%
|
| 60 |
+
MobileNetV5,Rock_Classification,INT8,NPU,59,453.605,83.33%
|
| 61 |
+
MobileNetV5,Rock_Classification,INT8,NPU,60,502.012,83.33%
|
| 62 |
+
MobileNetV5,Rock_Classification,INT8,NPU,61,432.019,83.33%
|
| 63 |
+
MobileNetV5,Rock_Classification,INT8,NPU,62,454.471,83.33%
|
| 64 |
+
MobileNetV5,Rock_Classification,INT8,NPU,63,440.109,83.33%
|
| 65 |
+
MobileNetV5,Rock_Classification,INT8,NPU,64,436.417,83.33%
|
| 66 |
+
MobileNetV5,Rock_Classification,INT8,NPU,65,445.619,83.33%
|
| 67 |
+
MobileNetV5,Rock_Classification,INT8,NPU,66,481.708,83.33%
|
| 68 |
+
MobileNetV5,Rock_Classification,INT8,NPU,67,445.686,83.33%
|
| 69 |
+
MobileNetV5,Rock_Classification,INT8,NPU,68,436.065,83.33%
|
| 70 |
+
MobileNetV5,Rock_Classification,INT8,NPU,69,437.401,83.33%
|
| 71 |
+
MobileNetV5,Rock_Classification,INT8,NPU,70,442.263,83.33%
|
| 72 |
+
MobileNetV5,Rock_Classification,INT8,NPU,71,451.373,83.33%
|
| 73 |
+
MobileNetV5,Rock_Classification,INT8,NPU,72,435.265,83.33%
|
| 74 |
+
MobileNetV5,Rock_Classification,INT8,NPU,73,482.138,83.33%
|
| 75 |
+
MobileNetV5,Rock_Classification,INT8,NPU,74,433.707,83.33%
|
| 76 |
+
MobileNetV5,Rock_Classification,INT8,NPU,75,437.502,83.33%
|
| 77 |
+
MobileNetV5,Rock_Classification,INT8,NPU,76,447.024,83.33%
|
| 78 |
+
MobileNetV5,Rock_Classification,INT8,NPU,77,454.967,83.33%
|
| 79 |
+
MobileNetV5,Rock_Classification,INT8,NPU,78,436.673,83.33%
|
| 80 |
+
MobileNetV5,Rock_Classification,INT8,NPU,79,450.949,83.33%
|
| 81 |
+
MobileNetV5,Rock_Classification,INT8,NPU,80,442.631,83.33%
|
| 82 |
+
MobileNetV5,Rock_Classification,INT8,NPU,81,444.245,83.33%
|
| 83 |
+
MobileNetV5,Rock_Classification,INT8,NPU,82,462.595,83.33%
|
| 84 |
+
MobileNetV5,Rock_Classification,INT8,NPU,83,448.490,83.33%
|
| 85 |
+
MobileNetV5,Rock_Classification,INT8,NPU,84,445.366,83.33%
|
| 86 |
+
MobileNetV5,Rock_Classification,INT8,NPU,85,447.867,83.33%
|
| 87 |
+
MobileNetV5,Rock_Classification,INT8,NPU,86,491.392,83.33%
|
| 88 |
+
MobileNetV5,Rock_Classification,INT8,NPU,87,439.673,83.33%
|
| 89 |
+
MobileNetV5,Rock_Classification,INT8,NPU,88,470.537,83.33%
|
| 90 |
+
MobileNetV5,Rock_Classification,INT8,NPU,89,448.300,83.33%
|
| 91 |
+
MobileNetV5,Rock_Classification,INT8,NPU,90,455.241,83.33%
|
| 92 |
+
MobileNetV5,Rock_Classification,INT8,NPU,91,448.663,83.33%
|
| 93 |
+
MobileNetV5,Rock_Classification,INT8,NPU,92,455.440,83.33%
|
| 94 |
+
MobileNetV5,Rock_Classification,INT8,NPU,93,439.700,83.33%
|
| 95 |
+
MobileNetV5,Rock_Classification,INT8,NPU,94,449.594,83.33%
|
| 96 |
+
MobileNetV5,Rock_Classification,INT8,NPU,95,446.067,83.33%
|
| 97 |
+
MobileNetV5,Rock_Classification,INT8,NPU,96,437.317,83.33%
|
| 98 |
+
MobileNetV5,Rock_Classification,INT8,NPU,97,449.063,83.33%
|
| 99 |
+
MobileNetV5,Rock_Classification,INT8,NPU,98,443.661,83.33%
|
| 100 |
+
MobileNetV5,Rock_Classification,INT8,NPU,99,491.462,83.33%
|
| 101 |
+
MobileNetV5,Rock_Classification,INT8,NPU,100,487.063,83.33%
|
| 102 |
+
MobileNetV5,Rock_Classification,INT8,CPU,1,4369.197,79.87%
|
| 103 |
+
MobileNetV5,Rock_Classification,INT8,CPU,2,4264.847,79.87%
|
| 104 |
+
MobileNetV5,Rock_Classification,INT8,CPU,3,4728.492,79.87%
|
| 105 |
+
MobileNetV5,Rock_Classification,INT8,CPU,4,4300.585,79.87%
|
| 106 |
+
MobileNetV5,Rock_Classification,INT8,CPU,5,3889.106,79.87%
|
| 107 |
+
MobileNetV5,Rock_Classification,INT8,CPU,6,4206.374,79.87%
|
| 108 |
+
MobileNetV5,Rock_Classification,INT8,CPU,7,4004.576,79.87%
|
| 109 |
+
MobileNetV5,Rock_Classification,INT8,CPU,8,4619.369,79.87%
|
| 110 |
+
MobileNetV5,Rock_Classification,INT8,CPU,9,4732.749,79.87%
|
| 111 |
+
MobileNetV5,Rock_Classification,INT8,CPU,10,4889.63,79.87%
|
| 112 |
+
MobileNetV5,Rock_Classification,INT8,CPU,11,4988.201,79.87%
|
| 113 |
+
MobileNetV5,Rock_Classification,INT8,CPU,12,4987.135,79.87%
|
| 114 |
+
MobileNetV5,Rock_Classification,INT8,CPU,13,4829.296,79.87%
|
| 115 |
+
MobileNetV5,Rock_Classification,INT8,CPU,14,4779.403,79.87%
|
| 116 |
+
MobileNetV5,Rock_Classification,INT8,CPU,15,4394.283,79.87%
|
| 117 |
+
MobileNetV5,Rock_Classification,INT8,CPU,16,4382.425,79.87%
|
| 118 |
+
MobileNetV5,Rock_Classification,INT8,CPU,17,4237.085,79.87%
|
| 119 |
+
MobileNetV5,Rock_Classification,INT8,CPU,18,5319.418,79.87%
|
| 120 |
+
MobileNetV5,Rock_Classification,INT8,CPU,19,4833.478,79.87%
|
| 121 |
+
MobileNetV5,Rock_Classification,INT8,CPU,20,3886.22,79.87%
|
| 122 |
+
MobileNetV5,Rock_Classification,INT8,CPU,21,4612.789,79.87%
|
| 123 |
+
MobileNetV5,Rock_Classification,INT8,CPU,22,5131.234,79.87%
|
| 124 |
+
MobileNetV5,Rock_Classification,INT8,CPU,23,3986.809,79.87%
|
| 125 |
+
MobileNetV5,Rock_Classification,INT8,CPU,24,3511.726,79.87%
|
| 126 |
+
MobileNetV5,Rock_Classification,INT8,CPU,25,5049.574,79.87%
|
| 127 |
+
MobileNetV5,Rock_Classification,INT8,CPU,26,5037.559,79.87%
|
| 128 |
+
MobileNetV5,Rock_Classification,INT8,CPU,27,4243.758,79.87%
|
| 129 |
+
MobileNetV5,Rock_Classification,INT8,CPU,28,4003.305,79.87%
|
| 130 |
+
MobileNetV5,Rock_Classification,INT8,CPU,29,3665.967,79.87%
|
| 131 |
+
MobileNetV5,Rock_Classification,INT8,CPU,30,4016.908,79.87%
|
| 132 |
+
MobileNetV5,Rock_Classification,INT8,CPU,31,4626.538,79.87%
|
| 133 |
+
MobileNetV5,Rock_Classification,INT8,CPU,32,4206.99,79.87%
|
| 134 |
+
MobileNetV5,Rock_Classification,INT8,CPU,33,4206.856,79.87%
|
| 135 |
+
MobileNetV5,Rock_Classification,INT8,CPU,34,4429.727,79.87%
|
| 136 |
+
MobileNetV5,Rock_Classification,INT8,CPU,35,3972.8,79.87%
|
| 137 |
+
MobileNetV5,Rock_Classification,INT8,CPU,36,4140.013,79.87%
|
| 138 |
+
MobileNetV5,Rock_Classification,INT8,CPU,37,3868.932,79.87%
|
| 139 |
+
MobileNetV5,Rock_Classification,INT8,CPU,38,3593.876,79.87%
|
| 140 |
+
MobileNetV5,Rock_Classification,INT8,CPU,39,4239.009,79.87%
|
| 141 |
+
MobileNetV5,Rock_Classification,INT8,CPU,40,3495.432,79.87%
|
| 142 |
+
MobileNetV5,Rock_Classification,INT8,CPU,41,4543.234,79.87%
|
| 143 |
+
MobileNetV5,Rock_Classification,INT8,CPU,42,4079.814,79.87%
|
| 144 |
+
MobileNetV5,Rock_Classification,INT8,CPU,43,4452.977,79.87%
|
| 145 |
+
MobileNetV5,Rock_Classification,INT8,CPU,44,4071.539,79.87%
|
| 146 |
+
MobileNetV5,Rock_Classification,INT8,CPU,45,4008.992,79.87%
|
| 147 |
+
MobileNetV5,Rock_Classification,INT8,CPU,46,3858.049,79.87%
|
| 148 |
+
MobileNetV5,Rock_Classification,INT8,CPU,47,4305.525,79.87%
|
| 149 |
+
MobileNetV5,Rock_Classification,INT8,CPU,48,4647.992,79.87%
|
| 150 |
+
MobileNetV5,Rock_Classification,INT8,CPU,49,4051.507,79.87%
|
| 151 |
+
MobileNetV5,Rock_Classification,INT8,CPU,50,3836.135,79.87%
|
| 152 |
+
MobileNetV5,Rock_Classification,INT8,CPU,51,4512.967,79.87%
|
| 153 |
+
MobileNetV5,Rock_Classification,INT8,CPU,52,4421.742,79.87%
|
| 154 |
+
MobileNetV5,Rock_Classification,INT8,CPU,53,4186.991,79.87%
|
| 155 |
+
MobileNetV5,Rock_Classification,INT8,CPU,54,4078.728,79.87%
|
| 156 |
+
MobileNetV5,Rock_Classification,INT8,CPU,55,4582.876,79.87%
|
| 157 |
+
MobileNetV5,Rock_Classification,INT8,CPU,56,4045.481,79.87%
|
| 158 |
+
MobileNetV5,Rock_Classification,INT8,CPU,57,4411.41,79.87%
|
| 159 |
+
MobileNetV5,Rock_Classification,INT8,CPU,58,4577.71,79.87%
|
| 160 |
+
MobileNetV5,Rock_Classification,INT8,CPU,59,4303.297,79.87%
|
| 161 |
+
MobileNetV5,Rock_Classification,INT8,CPU,60,4370.276,79.87%
|
| 162 |
+
MobileNetV5,Rock_Classification,INT8,CPU,61,4305.007,79.87%
|
| 163 |
+
MobileNetV5,Rock_Classification,INT8,CPU,62,5145.41,79.87%
|
| 164 |
+
MobileNetV5,Rock_Classification,INT8,CPU,63,3805.421,79.87%
|
| 165 |
+
MobileNetV5,Rock_Classification,INT8,CPU,64,3983.302,79.87%
|
| 166 |
+
MobileNetV5,Rock_Classification,INT8,CPU,65,4571.616,79.87%
|
| 167 |
+
MobileNetV5,Rock_Classification,INT8,CPU,66,4533.058,79.87%
|
| 168 |
+
MobileNetV5,Rock_Classification,INT8,CPU,67,3847.589,79.87%
|
| 169 |
+
MobileNetV5,Rock_Classification,INT8,CPU,68,4117.824,79.87%
|
| 170 |
+
MobileNetV5,Rock_Classification,INT8,CPU,69,3671.528,79.87%
|
| 171 |
+
MobileNetV5,Rock_Classification,INT8,CPU,70,3894.427,79.87%
|
| 172 |
+
MobileNetV5,Rock_Classification,INT8,CPU,71,3495.949,79.87%
|
| 173 |
+
MobileNetV5,Rock_Classification,INT8,CPU,72,4027.068,79.87%
|
| 174 |
+
MobileNetV5,Rock_Classification,INT8,CPU,73,4085.617,79.87%
|
| 175 |
+
MobileNetV5,Rock_Classification,INT8,CPU,74,3907.066,79.87%
|
| 176 |
+
MobileNetV5,Rock_Classification,INT8,CPU,75,4229.687,79.87%
|
| 177 |
+
MobileNetV5,Rock_Classification,INT8,CPU,76,3565.366,79.87%
|
| 178 |
+
MobileNetV5,Rock_Classification,INT8,CPU,77,4047.344,79.87%
|
| 179 |
+
MobileNetV5,Rock_Classification,INT8,CPU,78,4093.268,79.87%
|
| 180 |
+
MobileNetV5,Rock_Classification,INT8,CPU,79,3260.243,79.87%
|
| 181 |
+
MobileNetV5,Rock_Classification,INT8,CPU,80,4425.205,79.87%
|
| 182 |
+
MobileNetV5,Rock_Classification,INT8,CPU,81,3878.28,79.87%
|
| 183 |
+
MobileNetV5,Rock_Classification,INT8,CPU,82,4226.33,79.87%
|
| 184 |
+
MobileNetV5,Rock_Classification,INT8,CPU,83,3892.407,79.87%
|
| 185 |
+
MobileNetV5,Rock_Classification,INT8,CPU,84,3546.474,79.87%
|
| 186 |
+
MobileNetV5,Rock_Classification,INT8,CPU,85,3487.636,79.87%
|
| 187 |
+
MobileNetV5,Rock_Classification,INT8,CPU,86,3650.556,79.87%
|
| 188 |
+
MobileNetV5,Rock_Classification,INT8,CPU,87,3902.074,79.87%
|
| 189 |
+
MobileNetV5,Rock_Classification,INT8,CPU,88,3470.268,79.87%
|
| 190 |
+
MobileNetV5,Rock_Classification,INT8,CPU,89,3085.261,79.87%
|
| 191 |
+
MobileNetV5,Rock_Classification,INT8,CPU,90,2863.087,79.87%
|
| 192 |
+
MobileNetV5,Rock_Classification,INT8,CPU,91,3656.753,79.87%
|
| 193 |
+
MobileNetV5,Rock_Classification,INT8,CPU,92,3052.207,79.87%
|
| 194 |
+
MobileNetV5,Rock_Classification,INT8,CPU,93,3174.881,79.87%
|
| 195 |
+
MobileNetV5,Rock_Classification,INT8,CPU,94,3532.24,79.87%
|
| 196 |
+
MobileNetV5,Rock_Classification,INT8,CPU,95,2946.779,79.87%
|
| 197 |
+
MobileNetV5,Rock_Classification,INT8,CPU,96,3849.334,79.87%
|
| 198 |
+
MobileNetV5,Rock_Classification,INT8,CPU,97,5257.185,79.87%
|
| 199 |
+
MobileNetV5,Rock_Classification,INT8,CPU,98,3703.351,79.87%
|
| 200 |
+
MobileNetV5,Rock_Classification,INT8,CPU,99,3659.264,79.87%
|
| 201 |
+
MobileNetV5,Rock_Classification,INT8,CPU,100,3632.956,79.87%
|
| 202 |
+
MobileNetV5,Rock_Classification,INT8,GPU,1,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 203 |
+
MobileNetV5,Rock_Classification,INT8,GPU,2,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 204 |
+
MobileNetV5,Rock_Classification,INT8,GPU,3,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 205 |
+
MobileNetV5,Rock_Classification,INT8,GPU,4,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 206 |
+
MobileNetV5,Rock_Classification,INT8,GPU,5,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 207 |
+
MobileNetV5,Rock_Classification,INT8,GPU,6,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 208 |
+
MobileNetV5,Rock_Classification,INT8,GPU,7,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 209 |
+
MobileNetV5,Rock_Classification,INT8,GPU,8,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 210 |
+
MobileNetV5,Rock_Classification,INT8,GPU,9,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 211 |
+
MobileNetV5,Rock_Classification,INT8,GPU,10,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 212 |
+
MobileNetV5,Rock_Classification,INT8,GPU,11,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 213 |
+
MobileNetV5,Rock_Classification,INT8,GPU,12,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 214 |
+
MobileNetV5,Rock_Classification,INT8,GPU,13,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 215 |
+
MobileNetV5,Rock_Classification,INT8,GPU,14,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 216 |
+
MobileNetV5,Rock_Classification,INT8,GPU,15,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 217 |
+
MobileNetV5,Rock_Classification,INT8,GPU,16,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 218 |
+
MobileNetV5,Rock_Classification,INT8,GPU,17,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 219 |
+
MobileNetV5,Rock_Classification,INT8,GPU,18,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 220 |
+
MobileNetV5,Rock_Classification,INT8,GPU,19,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 221 |
+
MobileNetV5,Rock_Classification,INT8,GPU,20,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 222 |
+
MobileNetV5,Rock_Classification,INT8,GPU,21,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 223 |
+
MobileNetV5,Rock_Classification,INT8,GPU,22,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 224 |
+
MobileNetV5,Rock_Classification,INT8,GPU,23,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 225 |
+
MobileNetV5,Rock_Classification,INT8,GPU,24,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 226 |
+
MobileNetV5,Rock_Classification,INT8,GPU,25,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 227 |
+
MobileNetV5,Rock_Classification,INT8,GPU,26,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 228 |
+
MobileNetV5,Rock_Classification,INT8,GPU,27,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 229 |
+
MobileNetV5,Rock_Classification,INT8,GPU,28,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 230 |
+
MobileNetV5,Rock_Classification,INT8,GPU,29,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 231 |
+
MobileNetV5,Rock_Classification,INT8,GPU,30,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 232 |
+
MobileNetV5,Rock_Classification,INT8,GPU,31,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 233 |
+
MobileNetV5,Rock_Classification,INT8,GPU,32,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 234 |
+
MobileNetV5,Rock_Classification,INT8,GPU,33,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 235 |
+
MobileNetV5,Rock_Classification,INT8,GPU,34,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 236 |
+
MobileNetV5,Rock_Classification,INT8,GPU,35,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 237 |
+
MobileNetV5,Rock_Classification,INT8,GPU,36,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 238 |
+
MobileNetV5,Rock_Classification,INT8,GPU,37,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 239 |
+
MobileNetV5,Rock_Classification,INT8,GPU,38,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 240 |
+
MobileNetV5,Rock_Classification,INT8,GPU,39,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 241 |
+
MobileNetV5,Rock_Classification,INT8,GPU,40,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 242 |
+
MobileNetV5,Rock_Classification,INT8,GPU,41,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 243 |
+
MobileNetV5,Rock_Classification,INT8,GPU,42,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 244 |
+
MobileNetV5,Rock_Classification,INT8,GPU,43,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 245 |
+
MobileNetV5,Rock_Classification,INT8,GPU,44,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 246 |
+
MobileNetV5,Rock_Classification,INT8,GPU,45,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 247 |
+
MobileNetV5,Rock_Classification,INT8,GPU,46,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 248 |
+
MobileNetV5,Rock_Classification,INT8,GPU,47,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 249 |
+
MobileNetV5,Rock_Classification,INT8,GPU,48,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 250 |
+
MobileNetV5,Rock_Classification,INT8,GPU,49,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 251 |
+
MobileNetV5,Rock_Classification,INT8,GPU,50,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 252 |
+
MobileNetV5,Rock_Classification,INT8,GPU,51,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 253 |
+
MobileNetV5,Rock_Classification,INT8,GPU,52,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 254 |
+
MobileNetV5,Rock_Classification,INT8,GPU,53,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 255 |
+
MobileNetV5,Rock_Classification,INT8,GPU,54,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 256 |
+
MobileNetV5,Rock_Classification,INT8,GPU,55,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 257 |
+
MobileNetV5,Rock_Classification,INT8,GPU,56,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 258 |
+
MobileNetV5,Rock_Classification,INT8,GPU,57,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 259 |
+
MobileNetV5,Rock_Classification,INT8,GPU,58,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 260 |
+
MobileNetV5,Rock_Classification,INT8,GPU,59,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 261 |
+
MobileNetV5,Rock_Classification,INT8,GPU,60,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 262 |
+
MobileNetV5,Rock_Classification,INT8,GPU,61,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 263 |
+
MobileNetV5,Rock_Classification,INT8,GPU,62,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 264 |
+
MobileNetV5,Rock_Classification,INT8,GPU,63,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 265 |
+
MobileNetV5,Rock_Classification,INT8,GPU,64,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 266 |
+
MobileNetV5,Rock_Classification,INT8,GPU,65,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 267 |
+
MobileNetV5,Rock_Classification,INT8,GPU,66,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 268 |
+
MobileNetV5,Rock_Classification,INT8,GPU,67,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 269 |
+
MobileNetV5,Rock_Classification,INT8,GPU,68,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 270 |
+
MobileNetV5,Rock_Classification,INT8,GPU,69,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 271 |
+
MobileNetV5,Rock_Classification,INT8,GPU,70,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 272 |
+
MobileNetV5,Rock_Classification,INT8,GPU,71,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 273 |
+
MobileNetV5,Rock_Classification,INT8,GPU,72,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 274 |
+
MobileNetV5,Rock_Classification,INT8,GPU,73,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 275 |
+
MobileNetV5,Rock_Classification,INT8,GPU,74,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 276 |
+
MobileNetV5,Rock_Classification,INT8,GPU,75,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 277 |
+
MobileNetV5,Rock_Classification,INT8,GPU,76,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 278 |
+
MobileNetV5,Rock_Classification,INT8,GPU,77,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 279 |
+
MobileNetV5,Rock_Classification,INT8,GPU,78,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 280 |
+
MobileNetV5,Rock_Classification,INT8,GPU,79,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 281 |
+
MobileNetV5,Rock_Classification,INT8,GPU,80,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 282 |
+
MobileNetV5,Rock_Classification,INT8,GPU,81,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 283 |
+
MobileNetV5,Rock_Classification,INT8,GPU,82,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 284 |
+
MobileNetV5,Rock_Classification,INT8,GPU,83,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 285 |
+
MobileNetV5,Rock_Classification,INT8,GPU,84,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 286 |
+
MobileNetV5,Rock_Classification,INT8,GPU,85,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 287 |
+
MobileNetV5,Rock_Classification,INT8,GPU,86,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 288 |
+
MobileNetV5,Rock_Classification,INT8,GPU,87,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 289 |
+
MobileNetV5,Rock_Classification,INT8,GPU,88,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 290 |
+
MobileNetV5,Rock_Classification,INT8,GPU,89,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 291 |
+
MobileNetV5,Rock_Classification,INT8,GPU,90,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 292 |
+
MobileNetV5,Rock_Classification,INT8,GPU,91,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 293 |
+
MobileNetV5,Rock_Classification,INT8,GPU,92,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 294 |
+
MobileNetV5,Rock_Classification,INT8,GPU,93,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 295 |
+
MobileNetV5,Rock_Classification,INT8,GPU,94,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 296 |
+
MobileNetV5,Rock_Classification,INT8,GPU,95,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 297 |
+
MobileNetV5,Rock_Classification,INT8,GPU,96,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 298 |
+
MobileNetV5,Rock_Classification,INT8,GPU,97,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 299 |
+
MobileNetV5,Rock_Classification,INT8,GPU,98,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 300 |
+
MobileNetV5,Rock_Classification,INT8,GPU,99,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 301 |
+
MobileNetV5,Rock_Classification,INT8,GPU,100,N/A (INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 302 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,1,447.036,78.42%
|
| 303 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,2,573.522,78.42%
|
| 304 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,3,455.000,78.42%
|
| 305 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,4,481.350,78.42%
|
| 306 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,5,470.084,78.42%
|
| 307 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,6,495.132,78.42%
|
| 308 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,7,456.445,78.42%
|
| 309 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,8,455.894,78.42%
|
| 310 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,9,476.072,78.42%
|
| 311 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,10,446.283,78.42%
|
| 312 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,11,451.302,78.42%
|
| 313 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,12,462.856,78.42%
|
| 314 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,13,469.247,78.42%
|
| 315 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,14,466.349,78.42%
|
| 316 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,15,483.785,78.42%
|
| 317 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,16,486.223,78.42%
|
| 318 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,17,484.613,78.42%
|
| 319 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,18,477.915,78.42%
|
| 320 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,19,466.814,78.42%
|
| 321 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,20,489.633,78.42%
|
| 322 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,21,494.244,78.42%
|
| 323 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,22,483.990,78.42%
|
| 324 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,23,482.268,78.42%
|
| 325 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,24,499.809,78.42%
|
| 326 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,25,516.045,78.42%
|
| 327 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,26,468.968,78.42%
|
| 328 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,27,451.147,78.42%
|
| 329 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,28,461.940,78.42%
|
| 330 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,29,488.013,78.42%
|
| 331 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,30,480.316,78.42%
|
| 332 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,31,493.020,78.42%
|
| 333 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,32,479.907,78.42%
|
| 334 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,33,444.224,78.42%
|
| 335 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,34,446.395,78.42%
|
| 336 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,35,471.297,78.42%
|
| 337 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,36,481.173,78.42%
|
| 338 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,37,473.123,78.42%
|
| 339 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,38,499.735,78.42%
|
| 340 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,39,454.556,78.42%
|
| 341 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,40,461.861,78.42%
|
| 342 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,41,470.146,78.42%
|
| 343 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,42,496.696,78.42%
|
| 344 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,43,462.586,78.42%
|
| 345 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,44,476.813,78.42%
|
| 346 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,45,465.729,78.42%
|
| 347 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,46,483.743,78.42%
|
| 348 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,47,469.725,78.42%
|
| 349 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,48,442.606,78.42%
|
| 350 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,49,542.720,78.42%
|
| 351 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,50,467.032,78.42%
|
| 352 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,51,464.970,78.42%
|
| 353 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,52,468.866,78.42%
|
| 354 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,53,465.835,78.42%
|
| 355 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,54,459.256,78.42%
|
| 356 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,55,456.067,78.42%
|
| 357 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,56,498.737,78.42%
|
| 358 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,57,475.871,78.42%
|
| 359 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,58,461.316,78.42%
|
| 360 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,59,487.133,78.42%
|
| 361 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,60,484.397,78.42%
|
| 362 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,61,446.442,78.42%
|
| 363 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,62,490.458,78.42%
|
| 364 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,63,451.313,78.42%
|
| 365 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,64,439.930,78.42%
|
| 366 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,65,447.275,78.42%
|
| 367 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,66,508.244,78.42%
|
| 368 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,67,478.382,78.42%
|
| 369 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,68,491.861,78.42%
|
| 370 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,69,469.488,78.42%
|
| 371 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,70,485.822,78.42%
|
| 372 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,71,471.208,78.42%
|
| 373 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,72,478.083,78.42%
|
| 374 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,73,497.349,78.42%
|
| 375 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,74,474.805,78.42%
|
| 376 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,75,451.485,78.42%
|
| 377 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,76,550.227,78.42%
|
| 378 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,77,458.626,78.42%
|
| 379 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,78,476.268,78.42%
|
| 380 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,79,473.982,78.42%
|
| 381 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,80,474.529,78.42%
|
| 382 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,81,483.444,78.42%
|
| 383 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,82,560.748,78.42%
|
| 384 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,83,481.017,78.42%
|
| 385 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,84,486.676,78.42%
|
| 386 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,85,494.291,78.42%
|
| 387 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,86,485.792,78.42%
|
| 388 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,87,461.591,78.42%
|
| 389 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,88,481.035,78.42%
|
| 390 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,89,469.503,78.42%
|
| 391 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,90,469.011,78.42%
|
| 392 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,91,469.850,78.42%
|
| 393 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,92,478.769,78.42%
|
| 394 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,93,487.532,78.42%
|
| 395 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,94,463.178,78.42%
|
| 396 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,95,504.581,78.42%
|
| 397 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,96,502.763,78.42%
|
| 398 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,97,496.403,78.42%
|
| 399 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,98,475.971,78.42%
|
| 400 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,99,455.617,78.42%
|
| 401 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,100,446.729,78.42%
|
| 402 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,1,4886.189,78.42%
|
| 403 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,2,3776.318,78.42%
|
| 404 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,3,3658.063,78.42%
|
| 405 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,4,4584.301,78.42%
|
| 406 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,5,3737.988,78.42%
|
| 407 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,6,4053.131,78.42%
|
| 408 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,7,3741.874,78.42%
|
| 409 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,8,4003.182,78.42%
|
| 410 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,9,4112.545,78.42%
|
| 411 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,10,4009.532,78.42%
|
| 412 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,11,3567.013,78.42%
|
| 413 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,12,4558.658,78.42%
|
| 414 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,13,4065.828,78.42%
|
| 415 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,14,3830.591,78.42%
|
| 416 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,15,3881.24,78.42%
|
| 417 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,16,3665.999,78.42%
|
| 418 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,17,4147.233,78.42%
|
| 419 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,18,3688.149,78.42%
|
| 420 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,19,4003.41,78.42%
|
| 421 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,20,3537.753,78.42%
|
| 422 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,21,4033.115,78.42%
|
| 423 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,22,4028.053,78.42%
|
| 424 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,23,3938.217,78.42%
|
| 425 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,24,3982.91,78.42%
|
| 426 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,25,4132.661,78.42%
|
| 427 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,26,4198.893,78.42%
|
| 428 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,27,3850.805,78.42%
|
| 429 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,28,3691.342,78.42%
|
| 430 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,29,4392.667,78.42%
|
| 431 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,30,3864.66,78.42%
|
| 432 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,31,4263.355,78.42%
|
| 433 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,32,4582.538,78.42%
|
| 434 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,33,3978.541,78.42%
|
| 435 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,34,4511.994,78.42%
|
| 436 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,35,3714.972,78.42%
|
| 437 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,36,5207.848,78.42%
|
| 438 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,37,4803.458,78.42%
|
| 439 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,38,3946.811,78.42%
|
| 440 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,39,4106.679,78.42%
|
| 441 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,40,4022.156,78.42%
|
| 442 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,41,4095.228,78.42%
|
| 443 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,42,3908.313,78.42%
|
| 444 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,43,4470.521,78.42%
|
| 445 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,44,4768.087,78.42%
|
| 446 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,45,4391.277,78.42%
|
| 447 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,46,4266.932,78.42%
|
| 448 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,47,4318.603,78.42%
|
| 449 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,48,4460.442,78.42%
|
| 450 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,49,4077.548,78.42%
|
| 451 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,50,4430.826,78.42%
|
| 452 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,51,4469.186,78.42%
|
| 453 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,52,4394.823,78.42%
|
| 454 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,53,4458.327,78.42%
|
| 455 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,54,4669.857,78.42%
|
| 456 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,55,4658.265,78.42%
|
| 457 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,56,4429.788,78.42%
|
| 458 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,57,4540.284,78.42%
|
| 459 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,58,4116.821,78.42%
|
| 460 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,59,4135.337,78.42%
|
| 461 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,60,3884.616,78.42%
|
| 462 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,61,3298.768,78.42%
|
| 463 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,62,4153.387,78.42%
|
| 464 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,63,4140.197,78.42%
|
| 465 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,64,4727.063,78.42%
|
| 466 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,65,3889.635,78.42%
|
| 467 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,66,4089.986,78.42%
|
| 468 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,67,3975.033,78.42%
|
| 469 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,68,4795.395,78.42%
|
| 470 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,69,4301.348,78.42%
|
| 471 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,70,3971.95,78.42%
|
| 472 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,71,4605.282,78.42%
|
| 473 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,72,4472.462,78.42%
|
| 474 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,73,4071.866,78.42%
|
| 475 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,74,4121.98,78.42%
|
| 476 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,75,4399.047,78.42%
|
| 477 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,76,4504.252,78.42%
|
| 478 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,77,4571.637,78.42%
|
| 479 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,78,4493.498,78.42%
|
| 480 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,79,1243.539,78.42%
|
| 481 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,80,3867.48,78.42%
|
| 482 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,81,5253.67,78.42%
|
| 483 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,82,5337.246,78.42%
|
| 484 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,83,5238.533,78.42%
|
| 485 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,84,5418.453,78.42%
|
| 486 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,85,4768.792,78.42%
|
| 487 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,86,5026.502,78.42%
|
| 488 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,87,5679.012,78.42%
|
| 489 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,88,5063.292,78.42%
|
| 490 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,89,5097.979,78.42%
|
| 491 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,90,5858.533,78.42%
|
| 492 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,91,5106.229,78.42%
|
| 493 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,92,3471.535,78.42%
|
| 494 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,93,5088.69,78.42%
|
| 495 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,94,4858.794,78.42%
|
| 496 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,95,5028.264,78.42%
|
| 497 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,96,5089.579,78.42%
|
| 498 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,97,5369.288,78.42%
|
| 499 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,98,4859.758,78.42%
|
| 500 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,99,4721.001,78.42%
|
| 501 |
+
MobileNetV5,Rock_Classification,W4A8,CPU,100,4925.068,78.42%
|
| 502 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,1,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 503 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,2,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 504 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,3,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 505 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,4,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 506 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,5,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 507 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,6,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 508 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,7,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 509 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,8,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 510 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,9,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 511 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,10,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 512 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,11,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 513 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,12,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 514 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,13,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 515 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,14,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 516 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,15,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 517 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,16,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 518 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,17,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 519 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,18,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 520 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,19,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 521 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,20,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 522 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,21,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 523 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,22,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 524 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,23,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 525 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,24,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 526 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,25,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 527 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,26,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 528 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,27,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 529 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,28,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 530 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,29,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 531 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,30,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 532 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,31,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 533 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,32,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 534 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,33,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 535 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,34,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 536 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,35,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 537 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,36,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 538 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,37,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 539 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,38,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 540 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,39,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 541 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,40,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 542 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,41,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 543 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,42,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 544 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,43,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 545 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,44,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 546 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,45,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 547 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,46,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 548 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,47,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 549 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,48,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 550 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,49,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 551 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,50,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 552 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,51,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 553 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,52,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 554 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,53,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 555 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,54,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 556 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,55,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 557 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,56,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 558 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,57,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 559 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,58,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 560 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,59,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 561 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,60,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 562 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,61,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 563 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,62,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 564 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,63,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 565 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,64,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 566 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,65,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 567 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,66,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 568 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,67,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 569 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,68,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 570 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,69,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 571 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,70,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 572 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,71,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 573 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,72,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 574 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,73,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 575 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,74,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 576 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,75,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 577 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,76,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 578 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,77,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 579 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,78,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 580 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,79,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 581 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,80,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 582 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,81,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 583 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,82,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 584 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,83,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 585 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,84,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 586 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,85,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 587 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,86,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 588 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,87,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 589 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,88,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 590 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,89,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 591 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,90,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 592 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,91,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 593 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,92,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 594 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,93,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 595 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,94,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 596 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,95,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 597 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,96,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 598 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,97,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 599 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,98,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 600 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,99,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
| 601 |
+
MobileNetV5,Rock_Classification,W4A8,GPU,100,N/A (INT4/INT8 Op Unsupported by Adreno QnnGpu driver - FP16 only),N/A
|
results/MobileNetV5/machine_2/benchmark_machine_2_v5_comparative.csv
ADDED
|
@@ -0,0 +1,601 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Model,Task,Format,Backend,Inference_ID,Latency_ms,Accuracy
|
| 2 |
+
MobileNetV5,Rock_Classification,INT8,NPU,1,23.923,79.87%
|
| 3 |
+
MobileNetV5,Rock_Classification,INT8,NPU,2,25.056,79.87%
|
| 4 |
+
MobileNetV5,Rock_Classification,INT8,NPU,3,24.543,79.87%
|
| 5 |
+
MobileNetV5,Rock_Classification,INT8,NPU,4,24.759,79.87%
|
| 6 |
+
MobileNetV5,Rock_Classification,INT8,NPU,5,26.191,79.87%
|
| 7 |
+
MobileNetV5,Rock_Classification,INT8,NPU,6,24.499,79.87%
|
| 8 |
+
MobileNetV5,Rock_Classification,INT8,NPU,7,23.19,79.87%
|
| 9 |
+
MobileNetV5,Rock_Classification,INT8,NPU,8,22.883,79.87%
|
| 10 |
+
MobileNetV5,Rock_Classification,INT8,NPU,9,21.604,79.87%
|
| 11 |
+
MobileNetV5,Rock_Classification,INT8,NPU,10,23.774,79.87%
|
| 12 |
+
MobileNetV5,Rock_Classification,INT8,NPU,11,23.287,79.87%
|
| 13 |
+
MobileNetV5,Rock_Classification,INT8,NPU,12,22.237,79.87%
|
| 14 |
+
MobileNetV5,Rock_Classification,INT8,NPU,13,24.848,79.87%
|
| 15 |
+
MobileNetV5,Rock_Classification,INT8,NPU,14,25.212,79.87%
|
| 16 |
+
MobileNetV5,Rock_Classification,INT8,NPU,15,24.612,79.87%
|
| 17 |
+
MobileNetV5,Rock_Classification,INT8,NPU,16,24.431,79.87%
|
| 18 |
+
MobileNetV5,Rock_Classification,INT8,NPU,17,22.44,79.87%
|
| 19 |
+
MobileNetV5,Rock_Classification,INT8,NPU,18,25.957,79.87%
|
| 20 |
+
MobileNetV5,Rock_Classification,INT8,NPU,19,24.827,79.87%
|
| 21 |
+
MobileNetV5,Rock_Classification,INT8,NPU,20,24.094,79.87%
|
| 22 |
+
MobileNetV5,Rock_Classification,INT8,NPU,21,23.013,79.87%
|
| 23 |
+
MobileNetV5,Rock_Classification,INT8,NPU,22,24.387,79.87%
|
| 24 |
+
MobileNetV5,Rock_Classification,INT8,NPU,23,27.125,79.87%
|
| 25 |
+
MobileNetV5,Rock_Classification,INT8,NPU,24,23.793,79.87%
|
| 26 |
+
MobileNetV5,Rock_Classification,INT8,NPU,25,23.286,79.87%
|
| 27 |
+
MobileNetV5,Rock_Classification,INT8,NPU,26,23.145,79.87%
|
| 28 |
+
MobileNetV5,Rock_Classification,INT8,NPU,27,23.61,79.87%
|
| 29 |
+
MobileNetV5,Rock_Classification,INT8,NPU,28,22.957,79.87%
|
| 30 |
+
MobileNetV5,Rock_Classification,INT8,NPU,29,24.38,79.87%
|
| 31 |
+
MobileNetV5,Rock_Classification,INT8,NPU,30,23.732,79.87%
|
| 32 |
+
MobileNetV5,Rock_Classification,INT8,NPU,31,23.796,79.87%
|
| 33 |
+
MobileNetV5,Rock_Classification,INT8,NPU,32,22.894,79.87%
|
| 34 |
+
MobileNetV5,Rock_Classification,INT8,NPU,33,23.312,79.87%
|
| 35 |
+
MobileNetV5,Rock_Classification,INT8,NPU,34,23.841,79.87%
|
| 36 |
+
MobileNetV5,Rock_Classification,INT8,NPU,35,24.721,79.87%
|
| 37 |
+
MobileNetV5,Rock_Classification,INT8,NPU,36,23.821,79.87%
|
| 38 |
+
MobileNetV5,Rock_Classification,INT8,NPU,37,24.461,79.87%
|
| 39 |
+
MobileNetV5,Rock_Classification,INT8,NPU,38,23.335,79.87%
|
| 40 |
+
MobileNetV5,Rock_Classification,INT8,NPU,39,22.622,79.87%
|
| 41 |
+
MobileNetV5,Rock_Classification,INT8,NPU,40,23.732,79.87%
|
| 42 |
+
MobileNetV5,Rock_Classification,INT8,NPU,41,23.145,79.87%
|
| 43 |
+
MobileNetV5,Rock_Classification,INT8,NPU,42,25.178,79.87%
|
| 44 |
+
MobileNetV5,Rock_Classification,INT8,NPU,43,23.182,79.87%
|
| 45 |
+
MobileNetV5,Rock_Classification,INT8,NPU,44,25.192,79.87%
|
| 46 |
+
MobileNetV5,Rock_Classification,INT8,NPU,45,25.25,79.87%
|
| 47 |
+
MobileNetV5,Rock_Classification,INT8,NPU,46,23.279,79.87%
|
| 48 |
+
MobileNetV5,Rock_Classification,INT8,NPU,47,24.43,79.87%
|
| 49 |
+
MobileNetV5,Rock_Classification,INT8,NPU,48,24.307,79.87%
|
| 50 |
+
MobileNetV5,Rock_Classification,INT8,NPU,49,25.461,79.87%
|
| 51 |
+
MobileNetV5,Rock_Classification,INT8,NPU,50,23.134,79.87%
|
| 52 |
+
MobileNetV5,Rock_Classification,INT8,NPU,51,25.622,79.87%
|
| 53 |
+
MobileNetV5,Rock_Classification,INT8,NPU,52,24.085,79.87%
|
| 54 |
+
MobileNetV5,Rock_Classification,INT8,NPU,53,24.475,79.87%
|
| 55 |
+
MobileNetV5,Rock_Classification,INT8,NPU,54,23.676,79.87%
|
| 56 |
+
MobileNetV5,Rock_Classification,INT8,NPU,55,22.296,79.87%
|
| 57 |
+
MobileNetV5,Rock_Classification,INT8,NPU,56,22.785,79.87%
|
| 58 |
+
MobileNetV5,Rock_Classification,INT8,NPU,57,24.766,79.87%
|
| 59 |
+
MobileNetV5,Rock_Classification,INT8,NPU,58,23.584,79.87%
|
| 60 |
+
MobileNetV5,Rock_Classification,INT8,NPU,59,23.33,79.87%
|
| 61 |
+
MobileNetV5,Rock_Classification,INT8,NPU,60,25.984,79.87%
|
| 62 |
+
MobileNetV5,Rock_Classification,INT8,NPU,61,23.542,79.87%
|
| 63 |
+
MobileNetV5,Rock_Classification,INT8,NPU,62,23.239,79.87%
|
| 64 |
+
MobileNetV5,Rock_Classification,INT8,NPU,63,23.117,79.87%
|
| 65 |
+
MobileNetV5,Rock_Classification,INT8,NPU,64,24.646,79.87%
|
| 66 |
+
MobileNetV5,Rock_Classification,INT8,NPU,65,24.394,79.87%
|
| 67 |
+
MobileNetV5,Rock_Classification,INT8,NPU,66,24.949,79.87%
|
| 68 |
+
MobileNetV5,Rock_Classification,INT8,NPU,67,23.973,79.87%
|
| 69 |
+
MobileNetV5,Rock_Classification,INT8,NPU,68,22.812,79.87%
|
| 70 |
+
MobileNetV5,Rock_Classification,INT8,NPU,69,22.993,79.87%
|
| 71 |
+
MobileNetV5,Rock_Classification,INT8,NPU,70,24.2,79.87%
|
| 72 |
+
MobileNetV5,Rock_Classification,INT8,NPU,71,23.98,79.87%
|
| 73 |
+
MobileNetV5,Rock_Classification,INT8,NPU,72,24.044,79.87%
|
| 74 |
+
MobileNetV5,Rock_Classification,INT8,NPU,73,23.965,79.87%
|
| 75 |
+
MobileNetV5,Rock_Classification,INT8,NPU,74,24.524,79.87%
|
| 76 |
+
MobileNetV5,Rock_Classification,INT8,NPU,75,25.333,79.87%
|
| 77 |
+
MobileNetV5,Rock_Classification,INT8,NPU,76,25.03,79.87%
|
| 78 |
+
MobileNetV5,Rock_Classification,INT8,NPU,77,23.653,79.87%
|
| 79 |
+
MobileNetV5,Rock_Classification,INT8,NPU,78,24.7,79.87%
|
| 80 |
+
MobileNetV5,Rock_Classification,INT8,NPU,79,25.221,79.87%
|
| 81 |
+
MobileNetV5,Rock_Classification,INT8,NPU,80,23.693,79.87%
|
| 82 |
+
MobileNetV5,Rock_Classification,INT8,NPU,81,23.793,79.87%
|
| 83 |
+
MobileNetV5,Rock_Classification,INT8,NPU,82,25.227,79.87%
|
| 84 |
+
MobileNetV5,Rock_Classification,INT8,NPU,83,23.579,79.87%
|
| 85 |
+
MobileNetV5,Rock_Classification,INT8,NPU,84,23.521,79.87%
|
| 86 |
+
MobileNetV5,Rock_Classification,INT8,NPU,85,24.655,79.87%
|
| 87 |
+
MobileNetV5,Rock_Classification,INT8,NPU,86,22.915,79.87%
|
| 88 |
+
MobileNetV5,Rock_Classification,INT8,NPU,87,24.887,79.87%
|
| 89 |
+
MobileNetV5,Rock_Classification,INT8,NPU,88,24.924,79.87%
|
| 90 |
+
MobileNetV5,Rock_Classification,INT8,NPU,89,24.602,79.87%
|
| 91 |
+
MobileNetV5,Rock_Classification,INT8,NPU,90,23.177,79.87%
|
| 92 |
+
MobileNetV5,Rock_Classification,INT8,NPU,91,26.381,79.87%
|
| 93 |
+
MobileNetV5,Rock_Classification,INT8,NPU,92,24.745,79.87%
|
| 94 |
+
MobileNetV5,Rock_Classification,INT8,NPU,93,25.16,79.87%
|
| 95 |
+
MobileNetV5,Rock_Classification,INT8,NPU,94,24.845,79.87%
|
| 96 |
+
MobileNetV5,Rock_Classification,INT8,NPU,95,25.112,79.87%
|
| 97 |
+
MobileNetV5,Rock_Classification,INT8,NPU,96,25.676,79.87%
|
| 98 |
+
MobileNetV5,Rock_Classification,INT8,NPU,97,24.289,79.87%
|
| 99 |
+
MobileNetV5,Rock_Classification,INT8,NPU,98,25.425,79.87%
|
| 100 |
+
MobileNetV5,Rock_Classification,INT8,NPU,99,23.25,79.87%
|
| 101 |
+
MobileNetV5,Rock_Classification,INT8,NPU,100,24.119,79.87%
|
| 102 |
+
MobileNetV5,Rock_Classification,INT8,GPU,1,55.085,79.87%
|
| 103 |
+
MobileNetV5,Rock_Classification,INT8,GPU,2,53.681,79.87%
|
| 104 |
+
MobileNetV5,Rock_Classification,INT8,GPU,3,49.314,79.87%
|
| 105 |
+
MobileNetV5,Rock_Classification,INT8,GPU,4,53.814,79.87%
|
| 106 |
+
MobileNetV5,Rock_Classification,INT8,GPU,5,52.793,79.87%
|
| 107 |
+
MobileNetV5,Rock_Classification,INT8,GPU,6,54.371,79.87%
|
| 108 |
+
MobileNetV5,Rock_Classification,INT8,GPU,7,56.223,79.87%
|
| 109 |
+
MobileNetV5,Rock_Classification,INT8,GPU,8,54.961,79.87%
|
| 110 |
+
MobileNetV5,Rock_Classification,INT8,GPU,9,52.565,79.87%
|
| 111 |
+
MobileNetV5,Rock_Classification,INT8,GPU,10,52.095,79.87%
|
| 112 |
+
MobileNetV5,Rock_Classification,INT8,GPU,11,53.847,79.87%
|
| 113 |
+
MobileNetV5,Rock_Classification,INT8,GPU,12,55.519,79.87%
|
| 114 |
+
MobileNetV5,Rock_Classification,INT8,GPU,13,55.785,79.87%
|
| 115 |
+
MobileNetV5,Rock_Classification,INT8,GPU,14,55.499,79.87%
|
| 116 |
+
MobileNetV5,Rock_Classification,INT8,GPU,15,54.513,79.87%
|
| 117 |
+
MobileNetV5,Rock_Classification,INT8,GPU,16,53.698,79.87%
|
| 118 |
+
MobileNetV5,Rock_Classification,INT8,GPU,17,55.335,79.87%
|
| 119 |
+
MobileNetV5,Rock_Classification,INT8,GPU,18,57.519,79.87%
|
| 120 |
+
MobileNetV5,Rock_Classification,INT8,GPU,19,50.373,79.87%
|
| 121 |
+
MobileNetV5,Rock_Classification,INT8,GPU,20,56.768,79.87%
|
| 122 |
+
MobileNetV5,Rock_Classification,INT8,GPU,21,51.259,79.87%
|
| 123 |
+
MobileNetV5,Rock_Classification,INT8,GPU,22,54.238,79.87%
|
| 124 |
+
MobileNetV5,Rock_Classification,INT8,GPU,23,53.932,79.87%
|
| 125 |
+
MobileNetV5,Rock_Classification,INT8,GPU,24,53.522,79.87%
|
| 126 |
+
MobileNetV5,Rock_Classification,INT8,GPU,25,55.474,79.87%
|
| 127 |
+
MobileNetV5,Rock_Classification,INT8,GPU,26,56.551,79.87%
|
| 128 |
+
MobileNetV5,Rock_Classification,INT8,GPU,27,52.465,79.87%
|
| 129 |
+
MobileNetV5,Rock_Classification,INT8,GPU,28,53.986,79.87%
|
| 130 |
+
MobileNetV5,Rock_Classification,INT8,GPU,29,55.095,79.87%
|
| 131 |
+
MobileNetV5,Rock_Classification,INT8,GPU,30,50.954,79.87%
|
| 132 |
+
MobileNetV5,Rock_Classification,INT8,GPU,31,52.679,79.87%
|
| 133 |
+
MobileNetV5,Rock_Classification,INT8,GPU,32,52.861,79.87%
|
| 134 |
+
MobileNetV5,Rock_Classification,INT8,GPU,33,53.428,79.87%
|
| 135 |
+
MobileNetV5,Rock_Classification,INT8,GPU,34,56.417,79.87%
|
| 136 |
+
MobileNetV5,Rock_Classification,INT8,GPU,35,56.462,79.87%
|
| 137 |
+
MobileNetV5,Rock_Classification,INT8,GPU,36,52.469,79.87%
|
| 138 |
+
MobileNetV5,Rock_Classification,INT8,GPU,37,50.139,79.87%
|
| 139 |
+
MobileNetV5,Rock_Classification,INT8,GPU,38,55.871,79.87%
|
| 140 |
+
MobileNetV5,Rock_Classification,INT8,GPU,39,55.866,79.87%
|
| 141 |
+
MobileNetV5,Rock_Classification,INT8,GPU,40,54.45,79.87%
|
| 142 |
+
MobileNetV5,Rock_Classification,INT8,GPU,41,53.706,79.87%
|
| 143 |
+
MobileNetV5,Rock_Classification,INT8,GPU,42,56.035,79.87%
|
| 144 |
+
MobileNetV5,Rock_Classification,INT8,GPU,43,57.868,79.87%
|
| 145 |
+
MobileNetV5,Rock_Classification,INT8,GPU,44,54.612,79.87%
|
| 146 |
+
MobileNetV5,Rock_Classification,INT8,GPU,45,56.213,79.87%
|
| 147 |
+
MobileNetV5,Rock_Classification,INT8,GPU,46,51.372,79.87%
|
| 148 |
+
MobileNetV5,Rock_Classification,INT8,GPU,47,56.185,79.87%
|
| 149 |
+
MobileNetV5,Rock_Classification,INT8,GPU,48,54.749,79.87%
|
| 150 |
+
MobileNetV5,Rock_Classification,INT8,GPU,49,53.801,79.87%
|
| 151 |
+
MobileNetV5,Rock_Classification,INT8,GPU,50,54.69,79.87%
|
| 152 |
+
MobileNetV5,Rock_Classification,INT8,GPU,51,55.178,79.87%
|
| 153 |
+
MobileNetV5,Rock_Classification,INT8,GPU,52,60.613,79.87%
|
| 154 |
+
MobileNetV5,Rock_Classification,INT8,GPU,53,56.969,79.87%
|
| 155 |
+
MobileNetV5,Rock_Classification,INT8,GPU,54,57.519,79.87%
|
| 156 |
+
MobileNetV5,Rock_Classification,INT8,GPU,55,55.369,79.87%
|
| 157 |
+
MobileNetV5,Rock_Classification,INT8,GPU,56,53.779,79.87%
|
| 158 |
+
MobileNetV5,Rock_Classification,INT8,GPU,57,55.852,79.87%
|
| 159 |
+
MobileNetV5,Rock_Classification,INT8,GPU,58,53.923,79.87%
|
| 160 |
+
MobileNetV5,Rock_Classification,INT8,GPU,59,53.473,79.87%
|
| 161 |
+
MobileNetV5,Rock_Classification,INT8,GPU,60,56.644,79.87%
|
| 162 |
+
MobileNetV5,Rock_Classification,INT8,GPU,61,53.019,79.87%
|
| 163 |
+
MobileNetV5,Rock_Classification,INT8,GPU,62,56.113,79.87%
|
| 164 |
+
MobileNetV5,Rock_Classification,INT8,GPU,63,50.442,79.87%
|
| 165 |
+
MobileNetV5,Rock_Classification,INT8,GPU,64,51.279,79.87%
|
| 166 |
+
MobileNetV5,Rock_Classification,INT8,GPU,65,54.301,79.87%
|
| 167 |
+
MobileNetV5,Rock_Classification,INT8,GPU,66,51.49,79.87%
|
| 168 |
+
MobileNetV5,Rock_Classification,INT8,GPU,67,55.343,79.87%
|
| 169 |
+
MobileNetV5,Rock_Classification,INT8,GPU,68,54.144,79.87%
|
| 170 |
+
MobileNetV5,Rock_Classification,INT8,GPU,69,53.439,79.87%
|
| 171 |
+
MobileNetV5,Rock_Classification,INT8,GPU,70,53.11,79.87%
|
| 172 |
+
MobileNetV5,Rock_Classification,INT8,GPU,71,55.889,79.87%
|
| 173 |
+
MobileNetV5,Rock_Classification,INT8,GPU,72,55.024,79.87%
|
| 174 |
+
MobileNetV5,Rock_Classification,INT8,GPU,73,58.159,79.87%
|
| 175 |
+
MobileNetV5,Rock_Classification,INT8,GPU,74,57.503,79.87%
|
| 176 |
+
MobileNetV5,Rock_Classification,INT8,GPU,75,52.148,79.87%
|
| 177 |
+
MobileNetV5,Rock_Classification,INT8,GPU,76,55.9,79.87%
|
| 178 |
+
MobileNetV5,Rock_Classification,INT8,GPU,77,54.061,79.87%
|
| 179 |
+
MobileNetV5,Rock_Classification,INT8,GPU,78,52.437,79.87%
|
| 180 |
+
MobileNetV5,Rock_Classification,INT8,GPU,79,53.94,79.87%
|
| 181 |
+
MobileNetV5,Rock_Classification,INT8,GPU,80,55.306,79.87%
|
| 182 |
+
MobileNetV5,Rock_Classification,INT8,GPU,81,54.963,79.87%
|
| 183 |
+
MobileNetV5,Rock_Classification,INT8,GPU,82,51.048,79.87%
|
| 184 |
+
MobileNetV5,Rock_Classification,INT8,GPU,83,51.68,79.87%
|
| 185 |
+
MobileNetV5,Rock_Classification,INT8,GPU,84,52.935,79.87%
|
| 186 |
+
MobileNetV5,Rock_Classification,INT8,GPU,85,51.584,79.87%
|
| 187 |
+
MobileNetV5,Rock_Classification,INT8,GPU,86,54.201,79.87%
|
| 188 |
+
MobileNetV5,Rock_Classification,INT8,GPU,87,55.34,79.87%
|
| 189 |
+
MobileNetV5,Rock_Classification,INT8,GPU,88,53.899,79.87%
|
| 190 |
+
MobileNetV5,Rock_Classification,INT8,GPU,89,54.76,79.87%
|
| 191 |
+
MobileNetV5,Rock_Classification,INT8,GPU,90,53.002,79.87%
|
| 192 |
+
MobileNetV5,Rock_Classification,INT8,GPU,91,51.325,79.87%
|
| 193 |
+
MobileNetV5,Rock_Classification,INT8,GPU,92,54.231,79.87%
|
| 194 |
+
MobileNetV5,Rock_Classification,INT8,GPU,93,55.835,79.87%
|
| 195 |
+
MobileNetV5,Rock_Classification,INT8,GPU,94,54.931,79.87%
|
| 196 |
+
MobileNetV5,Rock_Classification,INT8,GPU,95,53.922,79.87%
|
| 197 |
+
MobileNetV5,Rock_Classification,INT8,GPU,96,55.056,79.87%
|
| 198 |
+
MobileNetV5,Rock_Classification,INT8,GPU,97,53.26,79.87%
|
| 199 |
+
MobileNetV5,Rock_Classification,INT8,GPU,98,51.564,79.87%
|
| 200 |
+
MobileNetV5,Rock_Classification,INT8,GPU,99,57.639,79.87%
|
| 201 |
+
MobileNetV5,Rock_Classification,INT8,GPU,100,54.539,79.87%
|
| 202 |
+
MobileNetV5,Rock_Classification,INT8,CPU,1,2570.365,79.87%
|
| 203 |
+
MobileNetV5,Rock_Classification,INT8,CPU,2,2498.693,79.87%
|
| 204 |
+
MobileNetV5,Rock_Classification,INT8,CPU,3,2489.67,79.87%
|
| 205 |
+
MobileNetV5,Rock_Classification,INT8,CPU,4,2495.03,79.87%
|
| 206 |
+
MobileNetV5,Rock_Classification,INT8,CPU,5,2410.581,79.87%
|
| 207 |
+
MobileNetV5,Rock_Classification,INT8,CPU,6,2415.569,79.87%
|
| 208 |
+
MobileNetV5,Rock_Classification,INT8,CPU,7,2614.949,79.87%
|
| 209 |
+
MobileNetV5,Rock_Classification,INT8,CPU,8,2551.237,79.87%
|
| 210 |
+
MobileNetV5,Rock_Classification,INT8,CPU,9,2336.941,79.87%
|
| 211 |
+
MobileNetV5,Rock_Classification,INT8,CPU,10,2496.797,79.87%
|
| 212 |
+
MobileNetV5,Rock_Classification,INT8,CPU,11,2531.928,79.87%
|
| 213 |
+
MobileNetV5,Rock_Classification,INT8,CPU,12,2517.233,79.87%
|
| 214 |
+
MobileNetV5,Rock_Classification,INT8,CPU,13,2460.633,79.87%
|
| 215 |
+
MobileNetV5,Rock_Classification,INT8,CPU,14,2488.981,79.87%
|
| 216 |
+
MobileNetV5,Rock_Classification,INT8,CPU,15,2526.064,79.87%
|
| 217 |
+
MobileNetV5,Rock_Classification,INT8,CPU,16,2421.755,79.87%
|
| 218 |
+
MobileNetV5,Rock_Classification,INT8,CPU,17,2444.316,79.87%
|
| 219 |
+
MobileNetV5,Rock_Classification,INT8,CPU,18,2434.207,79.87%
|
| 220 |
+
MobileNetV5,Rock_Classification,INT8,CPU,19,2566.947,79.87%
|
| 221 |
+
MobileNetV5,Rock_Classification,INT8,CPU,20,2407.971,79.87%
|
| 222 |
+
MobileNetV5,Rock_Classification,INT8,CPU,21,2360.392,79.87%
|
| 223 |
+
MobileNetV5,Rock_Classification,INT8,CPU,22,2285.861,79.87%
|
| 224 |
+
MobileNetV5,Rock_Classification,INT8,CPU,23,2581.875,79.87%
|
| 225 |
+
MobileNetV5,Rock_Classification,INT8,CPU,24,2498.704,79.87%
|
| 226 |
+
MobileNetV5,Rock_Classification,INT8,CPU,25,2502.028,79.87%
|
| 227 |
+
MobileNetV5,Rock_Classification,INT8,CPU,26,2456.057,79.87%
|
| 228 |
+
MobileNetV5,Rock_Classification,INT8,CPU,27,2557.694,79.87%
|
| 229 |
+
MobileNetV5,Rock_Classification,INT8,CPU,28,2701.692,79.87%
|
| 230 |
+
MobileNetV5,Rock_Classification,INT8,CPU,29,2353.182,79.87%
|
| 231 |
+
MobileNetV5,Rock_Classification,INT8,CPU,30,2420.082,79.87%
|
| 232 |
+
MobileNetV5,Rock_Classification,INT8,CPU,31,2584.382,79.87%
|
| 233 |
+
MobileNetV5,Rock_Classification,INT8,CPU,32,2393.991,79.87%
|
| 234 |
+
MobileNetV5,Rock_Classification,INT8,CPU,33,2390.761,79.87%
|
| 235 |
+
MobileNetV5,Rock_Classification,INT8,CPU,34,2449.26,79.87%
|
| 236 |
+
MobileNetV5,Rock_Classification,INT8,CPU,35,2482.649,79.87%
|
| 237 |
+
MobileNetV5,Rock_Classification,INT8,CPU,36,2405.136,79.87%
|
| 238 |
+
MobileNetV5,Rock_Classification,INT8,CPU,37,2566.417,79.87%
|
| 239 |
+
MobileNetV5,Rock_Classification,INT8,CPU,38,2267.583,79.87%
|
| 240 |
+
MobileNetV5,Rock_Classification,INT8,CPU,39,2505.009,79.87%
|
| 241 |
+
MobileNetV5,Rock_Classification,INT8,CPU,40,2248.145,79.87%
|
| 242 |
+
MobileNetV5,Rock_Classification,INT8,CPU,41,2459.758,79.87%
|
| 243 |
+
MobileNetV5,Rock_Classification,INT8,CPU,42,2346.059,79.87%
|
| 244 |
+
MobileNetV5,Rock_Classification,INT8,CPU,43,2437.039,79.87%
|
| 245 |
+
MobileNetV5,Rock_Classification,INT8,CPU,44,2438.633,79.87%
|
| 246 |
+
MobileNetV5,Rock_Classification,INT8,CPU,45,2434.411,79.87%
|
| 247 |
+
MobileNetV5,Rock_Classification,INT8,CPU,46,2466.169,79.87%
|
| 248 |
+
MobileNetV5,Rock_Classification,INT8,CPU,47,2480.727,79.87%
|
| 249 |
+
MobileNetV5,Rock_Classification,INT8,CPU,48,2439.583,79.87%
|
| 250 |
+
MobileNetV5,Rock_Classification,INT8,CPU,49,2530.041,79.87%
|
| 251 |
+
MobileNetV5,Rock_Classification,INT8,CPU,50,2455.631,79.87%
|
| 252 |
+
MobileNetV5,Rock_Classification,INT8,CPU,51,2396.096,79.87%
|
| 253 |
+
MobileNetV5,Rock_Classification,INT8,CPU,52,2310.139,79.87%
|
| 254 |
+
MobileNetV5,Rock_Classification,INT8,CPU,53,2340.991,79.87%
|
| 255 |
+
MobileNetV5,Rock_Classification,INT8,CPU,54,2336.794,79.87%
|
| 256 |
+
MobileNetV5,Rock_Classification,INT8,CPU,55,2380.695,79.87%
|
| 257 |
+
MobileNetV5,Rock_Classification,INT8,CPU,56,2404.004,79.87%
|
| 258 |
+
MobileNetV5,Rock_Classification,INT8,CPU,57,2584.997,79.87%
|
| 259 |
+
MobileNetV5,Rock_Classification,INT8,CPU,58,2452.635,79.87%
|
| 260 |
+
MobileNetV5,Rock_Classification,INT8,CPU,59,2512.593,79.87%
|
| 261 |
+
MobileNetV5,Rock_Classification,INT8,CPU,60,2401.438,79.87%
|
| 262 |
+
MobileNetV5,Rock_Classification,INT8,CPU,61,2601.728,79.87%
|
| 263 |
+
MobileNetV5,Rock_Classification,INT8,CPU,62,2468.424,79.87%
|
| 264 |
+
MobileNetV5,Rock_Classification,INT8,CPU,63,2422.913,79.87%
|
| 265 |
+
MobileNetV5,Rock_Classification,INT8,CPU,64,2437.682,79.87%
|
| 266 |
+
MobileNetV5,Rock_Classification,INT8,CPU,65,2463.155,79.87%
|
| 267 |
+
MobileNetV5,Rock_Classification,INT8,CPU,66,2447.616,79.87%
|
| 268 |
+
MobileNetV5,Rock_Classification,INT8,CPU,67,2392.518,79.87%
|
| 269 |
+
MobileNetV5,Rock_Classification,INT8,CPU,68,2497.97,79.87%
|
| 270 |
+
MobileNetV5,Rock_Classification,INT8,CPU,69,2723.179,79.87%
|
| 271 |
+
MobileNetV5,Rock_Classification,INT8,CPU,70,2608.648,79.87%
|
| 272 |
+
MobileNetV5,Rock_Classification,INT8,CPU,71,2562.986,79.87%
|
| 273 |
+
MobileNetV5,Rock_Classification,INT8,CPU,72,2357.016,79.87%
|
| 274 |
+
MobileNetV5,Rock_Classification,INT8,CPU,73,2385.448,79.87%
|
| 275 |
+
MobileNetV5,Rock_Classification,INT8,CPU,74,2331.891,79.87%
|
| 276 |
+
MobileNetV5,Rock_Classification,INT8,CPU,75,2473.485,79.87%
|
| 277 |
+
MobileNetV5,Rock_Classification,INT8,CPU,76,2423.707,79.87%
|
| 278 |
+
MobileNetV5,Rock_Classification,INT8,CPU,77,2538.412,79.87%
|
| 279 |
+
MobileNetV5,Rock_Classification,INT8,CPU,78,2401.798,79.87%
|
| 280 |
+
MobileNetV5,Rock_Classification,INT8,CPU,79,2488.98,79.87%
|
| 281 |
+
MobileNetV5,Rock_Classification,INT8,CPU,80,2537.647,79.87%
|
| 282 |
+
MobileNetV5,Rock_Classification,INT8,CPU,81,2446.944,79.87%
|
| 283 |
+
MobileNetV5,Rock_Classification,INT8,CPU,82,2358.825,79.87%
|
| 284 |
+
MobileNetV5,Rock_Classification,INT8,CPU,83,2495.342,79.87%
|
| 285 |
+
MobileNetV5,Rock_Classification,INT8,CPU,84,2345.196,79.87%
|
| 286 |
+
MobileNetV5,Rock_Classification,INT8,CPU,85,2417.77,79.87%
|
| 287 |
+
MobileNetV5,Rock_Classification,INT8,CPU,86,2427.37,79.87%
|
| 288 |
+
MobileNetV5,Rock_Classification,INT8,CPU,87,2443.662,79.87%
|
| 289 |
+
MobileNetV5,Rock_Classification,INT8,CPU,88,2447.766,79.87%
|
| 290 |
+
MobileNetV5,Rock_Classification,INT8,CPU,89,2395.401,79.87%
|
| 291 |
+
MobileNetV5,Rock_Classification,INT8,CPU,90,2454.55,79.87%
|
| 292 |
+
MobileNetV5,Rock_Classification,INT8,CPU,91,2405.97,79.87%
|
| 293 |
+
MobileNetV5,Rock_Classification,INT8,CPU,92,2395.231,79.87%
|
| 294 |
+
MobileNetV5,Rock_Classification,INT8,CPU,93,2523.634,79.87%
|
| 295 |
+
MobileNetV5,Rock_Classification,INT8,CPU,94,2457.696,79.87%
|
| 296 |
+
MobileNetV5,Rock_Classification,INT8,CPU,95,2373.992,79.87%
|
| 297 |
+
MobileNetV5,Rock_Classification,INT8,CPU,96,2512.418,79.87%
|
| 298 |
+
MobileNetV5,Rock_Classification,INT8,CPU,97,2298.319,79.87%
|
| 299 |
+
MobileNetV5,Rock_Classification,INT8,CPU,98,2562.201,79.87%
|
| 300 |
+
MobileNetV5,Rock_Classification,INT8,CPU,99,2224.824,79.87%
|
| 301 |
+
MobileNetV5,Rock_Classification,INT8,CPU,100,2645.162,79.87%
|
| 302 |
+
MobileNetV5,Rock_Classification,INT4,NPU,1,18.473,79.87%
|
| 303 |
+
MobileNetV5,Rock_Classification,INT4,NPU,2,19.579,79.87%
|
| 304 |
+
MobileNetV5,Rock_Classification,INT4,NPU,3,19.013,79.87%
|
| 305 |
+
MobileNetV5,Rock_Classification,INT4,NPU,4,18.433,79.87%
|
| 306 |
+
MobileNetV5,Rock_Classification,INT4,NPU,5,17.936,79.87%
|
| 307 |
+
MobileNetV5,Rock_Classification,INT4,NPU,6,17.872,79.87%
|
| 308 |
+
MobileNetV5,Rock_Classification,INT4,NPU,7,19.4,79.87%
|
| 309 |
+
MobileNetV5,Rock_Classification,INT4,NPU,8,17.452,79.87%
|
| 310 |
+
MobileNetV5,Rock_Classification,INT4,NPU,9,18.405,79.87%
|
| 311 |
+
MobileNetV5,Rock_Classification,INT4,NPU,10,19.418,79.87%
|
| 312 |
+
MobileNetV5,Rock_Classification,INT4,NPU,11,19.847,79.87%
|
| 313 |
+
MobileNetV5,Rock_Classification,INT4,NPU,12,18.515,79.87%
|
| 314 |
+
MobileNetV5,Rock_Classification,INT4,NPU,13,18.004,79.87%
|
| 315 |
+
MobileNetV5,Rock_Classification,INT4,NPU,14,17.687,79.87%
|
| 316 |
+
MobileNetV5,Rock_Classification,INT4,NPU,15,18.434,79.87%
|
| 317 |
+
MobileNetV5,Rock_Classification,INT4,NPU,16,16.992,79.87%
|
| 318 |
+
MobileNetV5,Rock_Classification,INT4,NPU,17,18.673,79.87%
|
| 319 |
+
MobileNetV5,Rock_Classification,INT4,NPU,18,18.742,79.87%
|
| 320 |
+
MobileNetV5,Rock_Classification,INT4,NPU,19,17.798,79.87%
|
| 321 |
+
MobileNetV5,Rock_Classification,INT4,NPU,20,18.572,79.87%
|
| 322 |
+
MobileNetV5,Rock_Classification,INT4,NPU,21,18.749,79.87%
|
| 323 |
+
MobileNetV5,Rock_Classification,INT4,NPU,22,18.629,79.87%
|
| 324 |
+
MobileNetV5,Rock_Classification,INT4,NPU,23,17.105,79.87%
|
| 325 |
+
MobileNetV5,Rock_Classification,INT4,NPU,24,18.721,79.87%
|
| 326 |
+
MobileNetV5,Rock_Classification,INT4,NPU,25,16.966,79.87%
|
| 327 |
+
MobileNetV5,Rock_Classification,INT4,NPU,26,18.187,79.87%
|
| 328 |
+
MobileNetV5,Rock_Classification,INT4,NPU,27,19.088,79.87%
|
| 329 |
+
MobileNetV5,Rock_Classification,INT4,NPU,28,17.84,79.87%
|
| 330 |
+
MobileNetV5,Rock_Classification,INT4,NPU,29,17.354,79.87%
|
| 331 |
+
MobileNetV5,Rock_Classification,INT4,NPU,30,18.135,79.87%
|
| 332 |
+
MobileNetV5,Rock_Classification,INT4,NPU,31,17.578,79.87%
|
| 333 |
+
MobileNetV5,Rock_Classification,INT4,NPU,32,19.14,79.87%
|
| 334 |
+
MobileNetV5,Rock_Classification,INT4,NPU,33,18.385,79.87%
|
| 335 |
+
MobileNetV5,Rock_Classification,INT4,NPU,34,17.583,79.87%
|
| 336 |
+
MobileNetV5,Rock_Classification,INT4,NPU,35,18.933,79.87%
|
| 337 |
+
MobileNetV5,Rock_Classification,INT4,NPU,36,19.782,79.87%
|
| 338 |
+
MobileNetV5,Rock_Classification,INT4,NPU,37,18.514,79.87%
|
| 339 |
+
MobileNetV5,Rock_Classification,INT4,NPU,38,17.443,79.87%
|
| 340 |
+
MobileNetV5,Rock_Classification,INT4,NPU,39,17.581,79.87%
|
| 341 |
+
MobileNetV5,Rock_Classification,INT4,NPU,40,18.966,79.87%
|
| 342 |
+
MobileNetV5,Rock_Classification,INT4,NPU,41,18.7,79.87%
|
| 343 |
+
MobileNetV5,Rock_Classification,INT4,NPU,42,18.66,79.87%
|
| 344 |
+
MobileNetV5,Rock_Classification,INT4,NPU,43,18.199,79.87%
|
| 345 |
+
MobileNetV5,Rock_Classification,INT4,NPU,44,19.359,79.87%
|
| 346 |
+
MobileNetV5,Rock_Classification,INT4,NPU,45,19.095,79.87%
|
| 347 |
+
MobileNetV5,Rock_Classification,INT4,NPU,46,19.572,79.87%
|
| 348 |
+
MobileNetV5,Rock_Classification,INT4,NPU,47,18.622,79.87%
|
| 349 |
+
MobileNetV5,Rock_Classification,INT4,NPU,48,18.414,79.87%
|
| 350 |
+
MobileNetV5,Rock_Classification,INT4,NPU,49,18.523,79.87%
|
| 351 |
+
MobileNetV5,Rock_Classification,INT4,NPU,50,18.357,79.87%
|
| 352 |
+
MobileNetV5,Rock_Classification,INT4,NPU,51,18.255,79.87%
|
| 353 |
+
MobileNetV5,Rock_Classification,INT4,NPU,52,18.144,79.87%
|
| 354 |
+
MobileNetV5,Rock_Classification,INT4,NPU,53,19.053,79.87%
|
| 355 |
+
MobileNetV5,Rock_Classification,INT4,NPU,54,17.343,79.87%
|
| 356 |
+
MobileNetV5,Rock_Classification,INT4,NPU,55,18.717,79.87%
|
| 357 |
+
MobileNetV5,Rock_Classification,INT4,NPU,56,19.539,79.87%
|
| 358 |
+
MobileNetV5,Rock_Classification,INT4,NPU,57,18.394,79.87%
|
| 359 |
+
MobileNetV5,Rock_Classification,INT4,NPU,58,18.976,79.87%
|
| 360 |
+
MobileNetV5,Rock_Classification,INT4,NPU,59,19.006,79.87%
|
| 361 |
+
MobileNetV5,Rock_Classification,INT4,NPU,60,18.34,79.87%
|
| 362 |
+
MobileNetV5,Rock_Classification,INT4,NPU,61,19.61,79.87%
|
| 363 |
+
MobileNetV5,Rock_Classification,INT4,NPU,62,18.991,79.87%
|
| 364 |
+
MobileNetV5,Rock_Classification,INT4,NPU,63,18.914,79.87%
|
| 365 |
+
MobileNetV5,Rock_Classification,INT4,NPU,64,17.862,79.87%
|
| 366 |
+
MobileNetV5,Rock_Classification,INT4,NPU,65,19.064,79.87%
|
| 367 |
+
MobileNetV5,Rock_Classification,INT4,NPU,66,18.838,79.87%
|
| 368 |
+
MobileNetV5,Rock_Classification,INT4,NPU,67,18.151,79.87%
|
| 369 |
+
MobileNetV5,Rock_Classification,INT4,NPU,68,19.19,79.87%
|
| 370 |
+
MobileNetV5,Rock_Classification,INT4,NPU,69,16.878,79.87%
|
| 371 |
+
MobileNetV5,Rock_Classification,INT4,NPU,70,18.558,79.87%
|
| 372 |
+
MobileNetV5,Rock_Classification,INT4,NPU,71,17.6,79.87%
|
| 373 |
+
MobileNetV5,Rock_Classification,INT4,NPU,72,17.66,79.87%
|
| 374 |
+
MobileNetV5,Rock_Classification,INT4,NPU,73,19.254,79.87%
|
| 375 |
+
MobileNetV5,Rock_Classification,INT4,NPU,74,17.92,79.87%
|
| 376 |
+
MobileNetV5,Rock_Classification,INT4,NPU,75,19.341,79.87%
|
| 377 |
+
MobileNetV5,Rock_Classification,INT4,NPU,76,17.973,79.87%
|
| 378 |
+
MobileNetV5,Rock_Classification,INT4,NPU,77,18.547,79.87%
|
| 379 |
+
MobileNetV5,Rock_Classification,INT4,NPU,78,18.553,79.87%
|
| 380 |
+
MobileNetV5,Rock_Classification,INT4,NPU,79,18.303,79.87%
|
| 381 |
+
MobileNetV5,Rock_Classification,INT4,NPU,80,18.71,79.87%
|
| 382 |
+
MobileNetV5,Rock_Classification,INT4,NPU,81,18.926,79.87%
|
| 383 |
+
MobileNetV5,Rock_Classification,INT4,NPU,82,18.443,79.87%
|
| 384 |
+
MobileNetV5,Rock_Classification,INT4,NPU,83,18.71,79.87%
|
| 385 |
+
MobileNetV5,Rock_Classification,INT4,NPU,84,17.396,79.87%
|
| 386 |
+
MobileNetV5,Rock_Classification,INT4,NPU,85,18.169,79.87%
|
| 387 |
+
MobileNetV5,Rock_Classification,INT4,NPU,86,19.693,79.87%
|
| 388 |
+
MobileNetV5,Rock_Classification,INT4,NPU,87,18.819,79.87%
|
| 389 |
+
MobileNetV5,Rock_Classification,INT4,NPU,88,18.008,79.87%
|
| 390 |
+
MobileNetV5,Rock_Classification,INT4,NPU,89,17.81,79.87%
|
| 391 |
+
MobileNetV5,Rock_Classification,INT4,NPU,90,18.533,79.87%
|
| 392 |
+
MobileNetV5,Rock_Classification,INT4,NPU,91,16.771,79.87%
|
| 393 |
+
MobileNetV5,Rock_Classification,INT4,NPU,92,17.994,79.87%
|
| 394 |
+
MobileNetV5,Rock_Classification,INT4,NPU,93,18.059,79.87%
|
| 395 |
+
MobileNetV5,Rock_Classification,INT4,NPU,94,18.199,79.87%
|
| 396 |
+
MobileNetV5,Rock_Classification,INT4,NPU,95,20.662,79.87%
|
| 397 |
+
MobileNetV5,Rock_Classification,INT4,NPU,96,17.855,79.87%
|
| 398 |
+
MobileNetV5,Rock_Classification,INT4,NPU,97,18.798,79.87%
|
| 399 |
+
MobileNetV5,Rock_Classification,INT4,NPU,98,17.632,79.87%
|
| 400 |
+
MobileNetV5,Rock_Classification,INT4,NPU,99,18.845,79.87%
|
| 401 |
+
MobileNetV5,Rock_Classification,INT4,NPU,100,18.671,79.87%
|
| 402 |
+
MobileNetV5,Rock_Classification,INT4,GPU,1,38.441,79.87%
|
| 403 |
+
MobileNetV5,Rock_Classification,INT4,GPU,2,37.978,79.87%
|
| 404 |
+
MobileNetV5,Rock_Classification,INT4,GPU,3,40.148,79.87%
|
| 405 |
+
MobileNetV5,Rock_Classification,INT4,GPU,4,39.679,79.87%
|
| 406 |
+
MobileNetV5,Rock_Classification,INT4,GPU,5,37.353,79.87%
|
| 407 |
+
MobileNetV5,Rock_Classification,INT4,GPU,6,39.252,79.87%
|
| 408 |
+
MobileNetV5,Rock_Classification,INT4,GPU,7,39.427,79.87%
|
| 409 |
+
MobileNetV5,Rock_Classification,INT4,GPU,8,38.276,79.87%
|
| 410 |
+
MobileNetV5,Rock_Classification,INT4,GPU,9,36.107,79.87%
|
| 411 |
+
MobileNetV5,Rock_Classification,INT4,GPU,10,37.708,79.87%
|
| 412 |
+
MobileNetV5,Rock_Classification,INT4,GPU,11,38.852,79.87%
|
| 413 |
+
MobileNetV5,Rock_Classification,INT4,GPU,12,38.763,79.87%
|
| 414 |
+
MobileNetV5,Rock_Classification,INT4,GPU,13,39.399,79.87%
|
| 415 |
+
MobileNetV5,Rock_Classification,INT4,GPU,14,39.683,79.87%
|
| 416 |
+
MobileNetV5,Rock_Classification,INT4,GPU,15,39.81,79.87%
|
| 417 |
+
MobileNetV5,Rock_Classification,INT4,GPU,16,38.935,79.87%
|
| 418 |
+
MobileNetV5,Rock_Classification,INT4,GPU,17,38.598,79.87%
|
| 419 |
+
MobileNetV5,Rock_Classification,INT4,GPU,18,41.984,79.87%
|
| 420 |
+
MobileNetV5,Rock_Classification,INT4,GPU,19,38.284,79.87%
|
| 421 |
+
MobileNetV5,Rock_Classification,INT4,GPU,20,36.713,79.87%
|
| 422 |
+
MobileNetV5,Rock_Classification,INT4,GPU,21,39.212,79.87%
|
| 423 |
+
MobileNetV5,Rock_Classification,INT4,GPU,22,41.077,79.87%
|
| 424 |
+
MobileNetV5,Rock_Classification,INT4,GPU,23,36.044,79.87%
|
| 425 |
+
MobileNetV5,Rock_Classification,INT4,GPU,24,38.667,79.87%
|
| 426 |
+
MobileNetV5,Rock_Classification,INT4,GPU,25,36.381,79.87%
|
| 427 |
+
MobileNetV5,Rock_Classification,INT4,GPU,26,39.183,79.87%
|
| 428 |
+
MobileNetV5,Rock_Classification,INT4,GPU,27,40.081,79.87%
|
| 429 |
+
MobileNetV5,Rock_Classification,INT4,GPU,28,40.352,79.87%
|
| 430 |
+
MobileNetV5,Rock_Classification,INT4,GPU,29,39.533,79.87%
|
| 431 |
+
MobileNetV5,Rock_Classification,INT4,GPU,30,38.237,79.87%
|
| 432 |
+
MobileNetV5,Rock_Classification,INT4,GPU,31,36.545,79.87%
|
| 433 |
+
MobileNetV5,Rock_Classification,INT4,GPU,32,38.276,79.87%
|
| 434 |
+
MobileNetV5,Rock_Classification,INT4,GPU,33,41.915,79.87%
|
| 435 |
+
MobileNetV5,Rock_Classification,INT4,GPU,34,37.566,79.87%
|
| 436 |
+
MobileNetV5,Rock_Classification,INT4,GPU,35,40.123,79.87%
|
| 437 |
+
MobileNetV5,Rock_Classification,INT4,GPU,36,37.692,79.87%
|
| 438 |
+
MobileNetV5,Rock_Classification,INT4,GPU,37,39.21,79.87%
|
| 439 |
+
MobileNetV5,Rock_Classification,INT4,GPU,38,40.544,79.87%
|
| 440 |
+
MobileNetV5,Rock_Classification,INT4,GPU,39,41.084,79.87%
|
| 441 |
+
MobileNetV5,Rock_Classification,INT4,GPU,40,38.068,79.87%
|
| 442 |
+
MobileNetV5,Rock_Classification,INT4,GPU,41,38.425,79.87%
|
| 443 |
+
MobileNetV5,Rock_Classification,INT4,GPU,42,37.245,79.87%
|
| 444 |
+
MobileNetV5,Rock_Classification,INT4,GPU,43,37.396,79.87%
|
| 445 |
+
MobileNetV5,Rock_Classification,INT4,GPU,44,41.814,79.87%
|
| 446 |
+
MobileNetV5,Rock_Classification,INT4,GPU,45,38.449,79.87%
|
| 447 |
+
MobileNetV5,Rock_Classification,INT4,GPU,46,38.777,79.87%
|
| 448 |
+
MobileNetV5,Rock_Classification,INT4,GPU,47,39.355,79.87%
|
| 449 |
+
MobileNetV5,Rock_Classification,INT4,GPU,48,37.081,79.87%
|
| 450 |
+
MobileNetV5,Rock_Classification,INT4,GPU,49,37.231,79.87%
|
| 451 |
+
MobileNetV5,Rock_Classification,INT4,GPU,50,39.403,79.87%
|
| 452 |
+
MobileNetV5,Rock_Classification,INT4,GPU,51,40.111,79.87%
|
| 453 |
+
MobileNetV5,Rock_Classification,INT4,GPU,52,37.786,79.87%
|
| 454 |
+
MobileNetV5,Rock_Classification,INT4,GPU,53,39.468,79.87%
|
| 455 |
+
MobileNetV5,Rock_Classification,INT4,GPU,54,39.458,79.87%
|
| 456 |
+
MobileNetV5,Rock_Classification,INT4,GPU,55,37.608,79.87%
|
| 457 |
+
MobileNetV5,Rock_Classification,INT4,GPU,56,40.482,79.87%
|
| 458 |
+
MobileNetV5,Rock_Classification,INT4,GPU,57,37.532,79.87%
|
| 459 |
+
MobileNetV5,Rock_Classification,INT4,GPU,58,40.971,79.87%
|
| 460 |
+
MobileNetV5,Rock_Classification,INT4,GPU,59,39.899,79.87%
|
| 461 |
+
MobileNetV5,Rock_Classification,INT4,GPU,60,40.488,79.87%
|
| 462 |
+
MobileNetV5,Rock_Classification,INT4,GPU,61,36.163,79.87%
|
| 463 |
+
MobileNetV5,Rock_Classification,INT4,GPU,62,37.645,79.87%
|
| 464 |
+
MobileNetV5,Rock_Classification,INT4,GPU,63,37.372,79.87%
|
| 465 |
+
MobileNetV5,Rock_Classification,INT4,GPU,64,41.126,79.87%
|
| 466 |
+
MobileNetV5,Rock_Classification,INT4,GPU,65,38.519,79.87%
|
| 467 |
+
MobileNetV5,Rock_Classification,INT4,GPU,66,39.094,79.87%
|
| 468 |
+
MobileNetV5,Rock_Classification,INT4,GPU,67,38.952,79.87%
|
| 469 |
+
MobileNetV5,Rock_Classification,INT4,GPU,68,38.369,79.87%
|
| 470 |
+
MobileNetV5,Rock_Classification,INT4,GPU,69,40.04,79.87%
|
| 471 |
+
MobileNetV5,Rock_Classification,INT4,GPU,70,40.119,79.87%
|
| 472 |
+
MobileNetV5,Rock_Classification,INT4,GPU,71,39.602,79.87%
|
| 473 |
+
MobileNetV5,Rock_Classification,INT4,GPU,72,39.045,79.87%
|
| 474 |
+
MobileNetV5,Rock_Classification,INT4,GPU,73,38.722,79.87%
|
| 475 |
+
MobileNetV5,Rock_Classification,INT4,GPU,74,38.811,79.87%
|
| 476 |
+
MobileNetV5,Rock_Classification,INT4,GPU,75,37.449,79.87%
|
| 477 |
+
MobileNetV5,Rock_Classification,INT4,GPU,76,39.513,79.87%
|
| 478 |
+
MobileNetV5,Rock_Classification,INT4,GPU,77,38.633,79.87%
|
| 479 |
+
MobileNetV5,Rock_Classification,INT4,GPU,78,38.066,79.87%
|
| 480 |
+
MobileNetV5,Rock_Classification,INT4,GPU,79,39.347,79.87%
|
| 481 |
+
MobileNetV5,Rock_Classification,INT4,GPU,80,41.006,79.87%
|
| 482 |
+
MobileNetV5,Rock_Classification,INT4,GPU,81,35.511,79.87%
|
| 483 |
+
MobileNetV5,Rock_Classification,INT4,GPU,82,39.314,79.87%
|
| 484 |
+
MobileNetV5,Rock_Classification,INT4,GPU,83,39.734,79.87%
|
| 485 |
+
MobileNetV5,Rock_Classification,INT4,GPU,84,38.635,79.87%
|
| 486 |
+
MobileNetV5,Rock_Classification,INT4,GPU,85,40.591,79.87%
|
| 487 |
+
MobileNetV5,Rock_Classification,INT4,GPU,86,37.908,79.87%
|
| 488 |
+
MobileNetV5,Rock_Classification,INT4,GPU,87,41.864,79.87%
|
| 489 |
+
MobileNetV5,Rock_Classification,INT4,GPU,88,41.697,79.87%
|
| 490 |
+
MobileNetV5,Rock_Classification,INT4,GPU,89,40.253,79.87%
|
| 491 |
+
MobileNetV5,Rock_Classification,INT4,GPU,90,35.727,79.87%
|
| 492 |
+
MobileNetV5,Rock_Classification,INT4,GPU,91,37.399,79.87%
|
| 493 |
+
MobileNetV5,Rock_Classification,INT4,GPU,92,37.964,79.87%
|
| 494 |
+
MobileNetV5,Rock_Classification,INT4,GPU,93,38.757,79.87%
|
| 495 |
+
MobileNetV5,Rock_Classification,INT4,GPU,94,38.596,79.87%
|
| 496 |
+
MobileNetV5,Rock_Classification,INT4,GPU,95,40.921,79.87%
|
| 497 |
+
MobileNetV5,Rock_Classification,INT4,GPU,96,39.784,79.87%
|
| 498 |
+
MobileNetV5,Rock_Classification,INT4,GPU,97,40.572,79.87%
|
| 499 |
+
MobileNetV5,Rock_Classification,INT4,GPU,98,37.883,79.87%
|
| 500 |
+
MobileNetV5,Rock_Classification,INT4,GPU,99,37.704,79.87%
|
| 501 |
+
MobileNetV5,Rock_Classification,INT4,GPU,100,42.177,79.87%
|
| 502 |
+
MobileNetV5,Rock_Classification,INT4,CPU,1,621.537,79.87%
|
| 503 |
+
MobileNetV5,Rock_Classification,INT4,CPU,2,620.654,79.87%
|
| 504 |
+
MobileNetV5,Rock_Classification,INT4,CPU,3,568.461,79.87%
|
| 505 |
+
MobileNetV5,Rock_Classification,INT4,CPU,4,580.875,79.87%
|
| 506 |
+
MobileNetV5,Rock_Classification,INT4,CPU,5,618.617,79.87%
|
| 507 |
+
MobileNetV5,Rock_Classification,INT4,CPU,6,591.289,79.87%
|
| 508 |
+
MobileNetV5,Rock_Classification,INT4,CPU,7,553.723,79.87%
|
| 509 |
+
MobileNetV5,Rock_Classification,INT4,CPU,8,577.917,79.87%
|
| 510 |
+
MobileNetV5,Rock_Classification,INT4,CPU,9,565.424,79.87%
|
| 511 |
+
MobileNetV5,Rock_Classification,INT4,CPU,10,579.597,79.87%
|
| 512 |
+
MobileNetV5,Rock_Classification,INT4,CPU,11,551.651,79.87%
|
| 513 |
+
MobileNetV5,Rock_Classification,INT4,CPU,12,594.007,79.87%
|
| 514 |
+
MobileNetV5,Rock_Classification,INT4,CPU,13,608.943,79.87%
|
| 515 |
+
MobileNetV5,Rock_Classification,INT4,CPU,14,583.92,79.87%
|
| 516 |
+
MobileNetV5,Rock_Classification,INT4,CPU,15,551.117,79.87%
|
| 517 |
+
MobileNetV5,Rock_Classification,INT4,CPU,16,610.967,79.87%
|
| 518 |
+
MobileNetV5,Rock_Classification,INT4,CPU,17,585.753,79.87%
|
| 519 |
+
MobileNetV5,Rock_Classification,INT4,CPU,18,598.316,79.87%
|
| 520 |
+
MobileNetV5,Rock_Classification,INT4,CPU,19,631.665,79.87%
|
| 521 |
+
MobileNetV5,Rock_Classification,INT4,CPU,20,576.233,79.87%
|
| 522 |
+
MobileNetV5,Rock_Classification,INT4,CPU,21,534.616,79.87%
|
| 523 |
+
MobileNetV5,Rock_Classification,INT4,CPU,22,599.645,79.87%
|
| 524 |
+
MobileNetV5,Rock_Classification,INT4,CPU,23,592.291,79.87%
|
| 525 |
+
MobileNetV5,Rock_Classification,INT4,CPU,24,587.014,79.87%
|
| 526 |
+
MobileNetV5,Rock_Classification,INT4,CPU,25,603.25,79.87%
|
| 527 |
+
MobileNetV5,Rock_Classification,INT4,CPU,26,553.813,79.87%
|
| 528 |
+
MobileNetV5,Rock_Classification,INT4,CPU,27,596.944,79.87%
|
| 529 |
+
MobileNetV5,Rock_Classification,INT4,CPU,28,558.894,79.87%
|
| 530 |
+
MobileNetV5,Rock_Classification,INT4,CPU,29,553.438,79.87%
|
| 531 |
+
MobileNetV5,Rock_Classification,INT4,CPU,30,545.918,79.87%
|
| 532 |
+
MobileNetV5,Rock_Classification,INT4,CPU,31,590.211,79.87%
|
| 533 |
+
MobileNetV5,Rock_Classification,INT4,CPU,32,590.467,79.87%
|
| 534 |
+
MobileNetV5,Rock_Classification,INT4,CPU,33,592.183,79.87%
|
| 535 |
+
MobileNetV5,Rock_Classification,INT4,CPU,34,608.388,79.87%
|
| 536 |
+
MobileNetV5,Rock_Classification,INT4,CPU,35,592.89,79.87%
|
| 537 |
+
MobileNetV5,Rock_Classification,INT4,CPU,36,599.515,79.87%
|
| 538 |
+
MobileNetV5,Rock_Classification,INT4,CPU,37,566.576,79.87%
|
| 539 |
+
MobileNetV5,Rock_Classification,INT4,CPU,38,649.938,79.87%
|
| 540 |
+
MobileNetV5,Rock_Classification,INT4,CPU,39,571.506,79.87%
|
| 541 |
+
MobileNetV5,Rock_Classification,INT4,CPU,40,599.781,79.87%
|
| 542 |
+
MobileNetV5,Rock_Classification,INT4,CPU,41,578.809,79.87%
|
| 543 |
+
MobileNetV5,Rock_Classification,INT4,CPU,42,578.207,79.87%
|
| 544 |
+
MobileNetV5,Rock_Classification,INT4,CPU,43,644.703,79.87%
|
| 545 |
+
MobileNetV5,Rock_Classification,INT4,CPU,44,581.79,79.87%
|
| 546 |
+
MobileNetV5,Rock_Classification,INT4,CPU,45,581.298,79.87%
|
| 547 |
+
MobileNetV5,Rock_Classification,INT4,CPU,46,619.89,79.87%
|
| 548 |
+
MobileNetV5,Rock_Classification,INT4,CPU,47,587.345,79.87%
|
| 549 |
+
MobileNetV5,Rock_Classification,INT4,CPU,48,597.416,79.87%
|
| 550 |
+
MobileNetV5,Rock_Classification,INT4,CPU,49,580.112,79.87%
|
| 551 |
+
MobileNetV5,Rock_Classification,INT4,CPU,50,608.583,79.87%
|
| 552 |
+
MobileNetV5,Rock_Classification,INT4,CPU,51,595.144,79.87%
|
| 553 |
+
MobileNetV5,Rock_Classification,INT4,CPU,52,580.57,79.87%
|
| 554 |
+
MobileNetV5,Rock_Classification,INT4,CPU,53,582.432,79.87%
|
| 555 |
+
MobileNetV5,Rock_Classification,INT4,CPU,54,588.827,79.87%
|
| 556 |
+
MobileNetV5,Rock_Classification,INT4,CPU,55,601.877,79.87%
|
| 557 |
+
MobileNetV5,Rock_Classification,INT4,CPU,56,569.655,79.87%
|
| 558 |
+
MobileNetV5,Rock_Classification,INT4,CPU,57,575.412,79.87%
|
| 559 |
+
MobileNetV5,Rock_Classification,INT4,CPU,58,606.76,79.87%
|
| 560 |
+
MobileNetV5,Rock_Classification,INT4,CPU,59,602.708,79.87%
|
| 561 |
+
MobileNetV5,Rock_Classification,INT4,CPU,60,575.98,79.87%
|
| 562 |
+
MobileNetV5,Rock_Classification,INT4,CPU,61,600.916,79.87%
|
| 563 |
+
MobileNetV5,Rock_Classification,INT4,CPU,62,574.514,79.87%
|
| 564 |
+
MobileNetV5,Rock_Classification,INT4,CPU,63,578.096,79.87%
|
| 565 |
+
MobileNetV5,Rock_Classification,INT4,CPU,64,592.13,79.87%
|
| 566 |
+
MobileNetV5,Rock_Classification,INT4,CPU,65,583.689,79.87%
|
| 567 |
+
MobileNetV5,Rock_Classification,INT4,CPU,66,558.196,79.87%
|
| 568 |
+
MobileNetV5,Rock_Classification,INT4,CPU,67,549.394,79.87%
|
| 569 |
+
MobileNetV5,Rock_Classification,INT4,CPU,68,577.476,79.87%
|
| 570 |
+
MobileNetV5,Rock_Classification,INT4,CPU,69,571.192,79.87%
|
| 571 |
+
MobileNetV5,Rock_Classification,INT4,CPU,70,587.817,79.87%
|
| 572 |
+
MobileNetV5,Rock_Classification,INT4,CPU,71,576.138,79.87%
|
| 573 |
+
MobileNetV5,Rock_Classification,INT4,CPU,72,609.379,79.87%
|
| 574 |
+
MobileNetV5,Rock_Classification,INT4,CPU,73,612.561,79.87%
|
| 575 |
+
MobileNetV5,Rock_Classification,INT4,CPU,74,629.419,79.87%
|
| 576 |
+
MobileNetV5,Rock_Classification,INT4,CPU,75,575.285,79.87%
|
| 577 |
+
MobileNetV5,Rock_Classification,INT4,CPU,76,561.205,79.87%
|
| 578 |
+
MobileNetV5,Rock_Classification,INT4,CPU,77,547.038,79.87%
|
| 579 |
+
MobileNetV5,Rock_Classification,INT4,CPU,78,606.031,79.87%
|
| 580 |
+
MobileNetV5,Rock_Classification,INT4,CPU,79,587.206,79.87%
|
| 581 |
+
MobileNetV5,Rock_Classification,INT4,CPU,80,588.259,79.87%
|
| 582 |
+
MobileNetV5,Rock_Classification,INT4,CPU,81,590.935,79.87%
|
| 583 |
+
MobileNetV5,Rock_Classification,INT4,CPU,82,555.738,79.87%
|
| 584 |
+
MobileNetV5,Rock_Classification,INT4,CPU,83,596.377,79.87%
|
| 585 |
+
MobileNetV5,Rock_Classification,INT4,CPU,84,575.037,79.87%
|
| 586 |
+
MobileNetV5,Rock_Classification,INT4,CPU,85,576.198,79.87%
|
| 587 |
+
MobileNetV5,Rock_Classification,INT4,CPU,86,554.995,79.87%
|
| 588 |
+
MobileNetV5,Rock_Classification,INT4,CPU,87,595.575,79.87%
|
| 589 |
+
MobileNetV5,Rock_Classification,INT4,CPU,88,554.577,79.87%
|
| 590 |
+
MobileNetV5,Rock_Classification,INT4,CPU,89,633.475,79.87%
|
| 591 |
+
MobileNetV5,Rock_Classification,INT4,CPU,90,502.319,79.87%
|
| 592 |
+
MobileNetV5,Rock_Classification,INT4,CPU,91,567.838,79.87%
|
| 593 |
+
MobileNetV5,Rock_Classification,INT4,CPU,92,555.413,79.87%
|
| 594 |
+
MobileNetV5,Rock_Classification,INT4,CPU,93,588.915,79.87%
|
| 595 |
+
MobileNetV5,Rock_Classification,INT4,CPU,94,590.904,79.87%
|
| 596 |
+
MobileNetV5,Rock_Classification,INT4,CPU,95,580.374,79.87%
|
| 597 |
+
MobileNetV5,Rock_Classification,INT4,CPU,96,598.337,79.87%
|
| 598 |
+
MobileNetV5,Rock_Classification,INT4,CPU,97,625.476,79.87%
|
| 599 |
+
MobileNetV5,Rock_Classification,INT4,CPU,98,582.164,79.87%
|
| 600 |
+
MobileNetV5,Rock_Classification,INT4,CPU,99,604.203,79.87%
|
| 601 |
+
MobileNetV5,Rock_Classification,INT4,CPU,100,563.078,79.87%
|
results/MobileNetV5/machine_2/benchmark_machine_2_v5_native_100_runs.csv
ADDED
|
@@ -0,0 +1,201 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Model,Task,Format,Backend,Inference_ID,Latency_ms,Top1_Accuracy,Top3_Accuracy,Hardware_Environment,Precision_Type,Execution_Status
|
| 2 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,1,2263.931,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 3 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,2,2089.384,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 4 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,3,2090.227,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 5 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,4,2151.333,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 6 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,5,2128.612,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 7 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,6,2051.719,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 8 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,7,2130.075,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 9 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,8,2024.949,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 10 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,9,1988.101,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 11 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,10,2050.399,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 12 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,11,1988.841,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 13 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,12,2009.28,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 14 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,13,2057.373,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 15 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,14,2134.655,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 16 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,15,2096.344,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 17 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,16,2148.523,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 18 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,17,2071.773,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 19 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,18,2137.718,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 20 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,19,2057.54,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 21 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,20,1979.736,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 22 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,21,2048.22,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 23 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,22,2038.429,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 24 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,23,2020.823,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 25 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,24,1984.131,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 26 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,25,2021.866,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 27 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,26,1999.332,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 28 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,27,1987.033,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 29 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,28,2014.682,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 30 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,29,1941.893,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 31 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,30,1956.975,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 32 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,31,2190.169,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 33 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,32,2354.139,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 34 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,33,1978.326,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 35 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,34,2057.904,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 36 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,35,1959.976,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 37 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,36,1952.763,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 38 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,37,2100.487,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 39 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,38,1979.421,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 40 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,39,1969.473,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 41 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,40,1981.318,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 42 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,41,1968.242,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 43 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,42,1956.995,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 44 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,43,2084.54,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 45 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,44,1992.212,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 46 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,45,1946.587,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 47 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,46,2077.397,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 48 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,47,1942.245,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 49 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,48,1950.874,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 50 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,49,2042.52,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 51 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,50,2005.699,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 52 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,51,1932.95,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 53 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,52,2011.773,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 54 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,53,1958.811,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 55 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,54,2098.531,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 56 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,55,2013.025,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 57 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,56,1991.057,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 58 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,57,2005.982,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 59 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,58,2008.459,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 60 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,59,1923.152,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 61 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,60,1931.056,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 62 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,61,2062.669,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 63 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,62,2035.127,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 64 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,63,1974.998,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 65 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,64,2042.916,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 66 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,65,1933.26,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 67 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,66,1965.817,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 68 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,67,2023.299,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 69 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,68,1944.277,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 70 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,69,1938.974,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 71 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,70,2095.878,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 72 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,71,2001.56,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 73 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,72,1960.689,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 74 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,73,2039.324,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 75 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,74,2059.531,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 76 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,75,2040.632,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 77 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,76,2003.702,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 78 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,77,1982.152,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 79 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,78,2012.821,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 80 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,79,2017.705,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 81 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,80,1919.116,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 82 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,81,1937.694,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 83 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,82,2036.279,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 84 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,83,1951.846,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 85 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,84,1976.694,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 86 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,85,1991.907,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 87 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,86,1910.184,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 88 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,87,1883.276,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 89 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,88,1972.358,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 90 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,89,1907.645,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 91 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,90,1958.026,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 92 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,91,1971.717,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 93 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,92,1976.807,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 94 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,93,1998.849,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 95 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,94,2009.96,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 96 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,95,1991.907,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 97 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,96,1979.167,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 98 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,97,2009.76,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 99 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,98,2001.709,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 100 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,99,2037.54,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 101 |
+
MobileNetV5,Rock_Classification,W4A8,NPU,100,2006.137,80.33%,80.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 102 |
+
MobileNetV5,Rock_Classification,INT8,NPU,1,2099.614,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 103 |
+
MobileNetV5,Rock_Classification,INT8,NPU,2,1973.06,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 104 |
+
MobileNetV5,Rock_Classification,INT8,NPU,3,1961.356,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 105 |
+
MobileNetV5,Rock_Classification,INT8,NPU,4,2028.792,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 106 |
+
MobileNetV5,Rock_Classification,INT8,NPU,5,1954.599,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 107 |
+
MobileNetV5,Rock_Classification,INT8,NPU,6,1974.646,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 108 |
+
MobileNetV5,Rock_Classification,INT8,NPU,7,2010.67,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 109 |
+
MobileNetV5,Rock_Classification,INT8,NPU,8,1943.781,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 110 |
+
MobileNetV5,Rock_Classification,INT8,NPU,9,1943.666,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 111 |
+
MobileNetV5,Rock_Classification,INT8,NPU,10,2011.241,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 112 |
+
MobileNetV5,Rock_Classification,INT8,NPU,11,1929.319,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 113 |
+
MobileNetV5,Rock_Classification,INT8,NPU,12,1956.03,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 114 |
+
MobileNetV5,Rock_Classification,INT8,NPU,13,2025.007,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 115 |
+
MobileNetV5,Rock_Classification,INT8,NPU,14,1974.267,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 116 |
+
MobileNetV5,Rock_Classification,INT8,NPU,15,1975.19,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 117 |
+
MobileNetV5,Rock_Classification,INT8,NPU,16,2028.191,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 118 |
+
MobileNetV5,Rock_Classification,INT8,NPU,17,1912.953,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 119 |
+
MobileNetV5,Rock_Classification,INT8,NPU,18,1996.448,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 120 |
+
MobileNetV5,Rock_Classification,INT8,NPU,19,2055.711,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 121 |
+
MobileNetV5,Rock_Classification,INT8,NPU,20,1932.627,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 122 |
+
MobileNetV5,Rock_Classification,INT8,NPU,21,1968.438,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 123 |
+
MobileNetV5,Rock_Classification,INT8,NPU,22,2049.779,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 124 |
+
MobileNetV5,Rock_Classification,INT8,NPU,23,1989.971,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 125 |
+
MobileNetV5,Rock_Classification,INT8,NPU,24,1993.116,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 126 |
+
MobileNetV5,Rock_Classification,INT8,NPU,25,2011.654,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 127 |
+
MobileNetV5,Rock_Classification,INT8,NPU,26,1933.573,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 128 |
+
MobileNetV5,Rock_Classification,INT8,NPU,27,1946.918,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 129 |
+
MobileNetV5,Rock_Classification,INT8,NPU,28,2069.834,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 130 |
+
MobileNetV5,Rock_Classification,INT8,NPU,29,2026.352,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 131 |
+
MobileNetV5,Rock_Classification,INT8,NPU,30,1992.82,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 132 |
+
MobileNetV5,Rock_Classification,INT8,NPU,31,2064.882,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 133 |
+
MobileNetV5,Rock_Classification,INT8,NPU,32,1928.199,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 134 |
+
MobileNetV5,Rock_Classification,INT8,NPU,33,1966.363,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 135 |
+
MobileNetV5,Rock_Classification,INT8,NPU,34,2019.631,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 136 |
+
MobileNetV5,Rock_Classification,INT8,NPU,35,1968.271,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 137 |
+
MobileNetV5,Rock_Classification,INT8,NPU,36,1939.039,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 138 |
+
MobileNetV5,Rock_Classification,INT8,NPU,37,2008.639,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 139 |
+
MobileNetV5,Rock_Classification,INT8,NPU,38,1956.799,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 140 |
+
MobileNetV5,Rock_Classification,INT8,NPU,39,1937.036,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 141 |
+
MobileNetV5,Rock_Classification,INT8,NPU,40,2007.364,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 142 |
+
MobileNetV5,Rock_Classification,INT8,NPU,41,1943.403,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 143 |
+
MobileNetV5,Rock_Classification,INT8,NPU,42,1949.679,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 144 |
+
MobileNetV5,Rock_Classification,INT8,NPU,43,2046.608,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 145 |
+
MobileNetV5,Rock_Classification,INT8,NPU,44,1996.151,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 146 |
+
MobileNetV5,Rock_Classification,INT8,NPU,45,1994.197,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 147 |
+
MobileNetV5,Rock_Classification,INT8,NPU,46,2035.304,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 148 |
+
MobileNetV5,Rock_Classification,INT8,NPU,47,1975.488,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 149 |
+
MobileNetV5,Rock_Classification,INT8,NPU,48,1944.53,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 150 |
+
MobileNetV5,Rock_Classification,INT8,NPU,49,2029.034,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 151 |
+
MobileNetV5,Rock_Classification,INT8,NPU,50,1994.18,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 152 |
+
MobileNetV5,Rock_Classification,INT8,NPU,51,1919.742,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 153 |
+
MobileNetV5,Rock_Classification,INT8,NPU,52,1982.446,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 154 |
+
MobileNetV5,Rock_Classification,INT8,NPU,53,1958.022,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 155 |
+
MobileNetV5,Rock_Classification,INT8,NPU,54,1951.584,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 156 |
+
MobileNetV5,Rock_Classification,INT8,NPU,55,2016.801,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 157 |
+
MobileNetV5,Rock_Classification,INT8,NPU,56,1944.939,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 158 |
+
MobileNetV5,Rock_Classification,INT8,NPU,57,1930.179,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 159 |
+
MobileNetV5,Rock_Classification,INT8,NPU,58,2040.967,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 160 |
+
MobileNetV5,Rock_Classification,INT8,NPU,59,1973.602,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 161 |
+
MobileNetV5,Rock_Classification,INT8,NPU,60,1985.825,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 162 |
+
MobileNetV5,Rock_Classification,INT8,NPU,61,2029.393,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 163 |
+
MobileNetV5,Rock_Classification,INT8,NPU,62,1936.405,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 164 |
+
MobileNetV5,Rock_Classification,INT8,NPU,63,1886.461,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 165 |
+
MobileNetV5,Rock_Classification,INT8,NPU,64,1999.662,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 166 |
+
MobileNetV5,Rock_Classification,INT8,NPU,65,1959.379,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 167 |
+
MobileNetV5,Rock_Classification,INT8,NPU,66,1928.269,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 168 |
+
MobileNetV5,Rock_Classification,INT8,NPU,67,2012.185,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 169 |
+
MobileNetV5,Rock_Classification,INT8,NPU,68,2003.072,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 170 |
+
MobileNetV5,Rock_Classification,INT8,NPU,69,1931.738,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 171 |
+
MobileNetV5,Rock_Classification,INT8,NPU,70,2003.986,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 172 |
+
MobileNetV5,Rock_Classification,INT8,NPU,71,1864.093,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 173 |
+
MobileNetV5,Rock_Classification,INT8,NPU,72,1932.241,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 174 |
+
MobileNetV5,Rock_Classification,INT8,NPU,73,1984.216,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 175 |
+
MobileNetV5,Rock_Classification,INT8,NPU,74,1885.218,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 176 |
+
MobileNetV5,Rock_Classification,INT8,NPU,75,1901.327,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 177 |
+
MobileNetV5,Rock_Classification,INT8,NPU,76,2021.82,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 178 |
+
MobileNetV5,Rock_Classification,INT8,NPU,77,1959.202,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 179 |
+
MobileNetV5,Rock_Classification,INT8,NPU,78,1902.883,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 180 |
+
MobileNetV5,Rock_Classification,INT8,NPU,79,2013.852,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 181 |
+
MobileNetV5,Rock_Classification,INT8,NPU,80,1956.683,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 182 |
+
MobileNetV5,Rock_Classification,INT8,NPU,81,2002.913,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 183 |
+
MobileNetV5,Rock_Classification,INT8,NPU,82,2027.971,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 184 |
+
MobileNetV5,Rock_Classification,INT8,NPU,83,1937.647,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 185 |
+
MobileNetV5,Rock_Classification,INT8,NPU,84,1914.906,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 186 |
+
MobileNetV5,Rock_Classification,INT8,NPU,85,1986.069,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 187 |
+
MobileNetV5,Rock_Classification,INT8,NPU,86,1934.313,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 188 |
+
MobileNetV5,Rock_Classification,INT8,NPU,87,1936.267,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 189 |
+
MobileNetV5,Rock_Classification,INT8,NPU,88,1995.703,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 190 |
+
MobileNetV5,Rock_Classification,INT8,NPU,89,1909.094,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 191 |
+
MobileNetV5,Rock_Classification,INT8,NPU,90,1988.872,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 192 |
+
MobileNetV5,Rock_Classification,INT8,NPU,91,2030.178,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 193 |
+
MobileNetV5,Rock_Classification,INT8,NPU,92,1957.505,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 194 |
+
MobileNetV5,Rock_Classification,INT8,NPU,93,1960.375,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 195 |
+
MobileNetV5,Rock_Classification,INT8,NPU,94,2055.535,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 196 |
+
MobileNetV5,Rock_Classification,INT8,NPU,95,1936.034,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 197 |
+
MobileNetV5,Rock_Classification,INT8,NPU,96,1903.725,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 198 |
+
MobileNetV5,Rock_Classification,INT8,NPU,97,1948.44,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 199 |
+
MobileNetV5,Rock_Classification,INT8,NPU,98,1913.244,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 200 |
+
MobileNetV5,Rock_Classification,INT8,NPU,99,1918.057,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
| 201 |
+
MobileNetV5,Rock_Classification,INT8,NPU,100,1974.076,83.33%,83.33%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated
|
results/MobileNetV5/machine_2/mobilenetv5_machine_2_executive_dashboard.png
ADDED
|
results/MobileNetV5/machine_2/mobilenetv5_machine_2_hardware_latency_comparison.png
ADDED
|
results/MobileNetV5/machine_2/mobilenetv5_machine_2_latency_comparison.png
ADDED
|
results/MobileNetV5/machine_2/mobilenetv5_machine_2_native_stability.png
ADDED
|
results/MobileNetV5/machine_2/mobilenetv5_machine_2_npu_stability_distribution.png
ADDED
|
results/MobileNetV5/machine_2/mobilenetv5_machine_2_pareto_accuracy_vs_latency.png
ADDED
|
results/MobileNetV5/machine_2/mobilenetv5_machine_2_quantization_paradigm_comparison.png
ADDED
|
results/MobileNetV5/machine_2/mobilenetv5_machine_2_stability_100_runs.png
ADDED
|
results/ScaleDetection/machine_1/benchmark_machine_1_scale.csv
ADDED
|
@@ -0,0 +1,401 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Model,Task,Format,Backend,Inference_ID,Latency_ms,Empirical_Accuracy,Correct_Predictions,Precision_Type,Active_Provider,Hardware_Environment
|
| 2 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,1,8.394,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 3 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,2,10.344,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 4 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,3,10.488,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 5 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,4,9.991,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 6 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,5,9.602,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 7 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,6,9.468,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 8 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,7,9.611,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 9 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,8,10.068,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 10 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,9,10.042,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 11 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,10,9.578,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 12 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,11,9.096,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 13 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,12,9.132,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 14 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,13,8.834,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 15 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,14,9.642,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 16 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,15,9.006,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 17 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,16,10.283,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 18 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,17,12.788,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 19 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,18,11.416,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 20 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,19,8.531,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 21 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,20,15.53,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 22 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,21,14.073,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 23 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,22,13.0,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 24 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,23,14.173,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 25 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,24,12.326,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 26 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,25,57.172,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 27 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,26,11.142,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 28 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,27,12.568,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 29 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,28,8.725,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 30 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,29,8.87,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 31 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,30,10.104,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 32 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,31,10.129,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 33 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,32,10.464,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 34 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,33,9.655,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 35 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,34,9.994,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 36 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,35,12.099,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 37 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,36,10.619,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 38 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,37,12.16,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 39 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,38,9.287,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 40 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,39,11.148,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 41 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,40,10.33,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 42 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,41,9.168,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 43 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,42,10.292,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 44 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,43,9.774,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 45 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,44,8.75,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 46 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,45,12.824,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 47 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,46,9.381,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 48 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,47,8.693,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 49 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,48,9.735,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 50 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,49,8.558,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 51 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,50,8.483,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 52 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,51,9.806,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 53 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,52,9.147,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 54 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,53,10.139,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 55 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,54,11.119,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 56 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,55,10.014,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 57 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,56,9.008,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 58 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,57,11.687,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 59 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,58,8.865,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 60 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,59,8.426,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 61 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,60,8.648,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 62 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,61,9.586,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 63 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,62,8.948,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 64 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,63,9.235,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 65 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,64,8.392,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 66 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,65,9.303,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 67 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,66,7.638,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 68 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,67,10.243,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 69 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,68,9.329,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 70 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,69,12.202,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 71 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,70,10.954,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 72 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,71,10.092,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 73 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,72,10.921,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 74 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,73,8.659,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 75 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,74,8.7,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 76 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,75,9.86,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 77 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,76,8.335,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 78 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,77,8.276,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 79 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,78,10.201,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 80 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,79,9.024,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 81 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,80,9.892,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 82 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,81,9.702,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 83 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,82,9.261,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 84 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,83,9.278,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 85 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,84,8.782,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 86 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,85,8.757,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 87 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,86,11.612,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 88 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,87,9.133,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 89 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,88,8.752,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 90 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,89,11.627,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 91 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,90,10.812,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 92 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,91,8.628,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 93 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,92,8.832,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 94 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,93,12.683,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 95 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,94,12.345,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 96 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,95,8.994,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 97 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,96,7.553,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 98 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,97,8.204,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 99 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,98,7.651,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 100 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,99,9.238,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 101 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,GPU_RTX5060,100,12.681,96.88%,93/96,INT8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 102 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,1,8.358,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 103 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,2,9.782,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 104 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,3,8.609,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 105 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,4,9.293,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 106 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,5,8.111,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 107 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,6,9.705,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 108 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,7,9.164,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 109 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,8,10.15,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 110 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,9,10.319,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 111 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,10,9.071,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 112 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,11,9.703,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 113 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,12,9.006,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 114 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,13,9.973,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 115 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,14,10.868,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 116 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,15,8.818,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 117 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,16,9.709,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 118 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,17,9.358,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 119 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,18,10.024,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 120 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,19,11.881,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 121 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,20,10.624,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 122 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,21,10.258,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 123 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,22,8.192,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 124 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,23,9.867,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 125 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,24,8.583,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 126 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,25,8.849,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 127 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,26,9.416,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 128 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,27,8.607,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 129 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,28,9.55,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 130 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,29,8.662,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 131 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,30,10.288,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 132 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,31,9.501,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 133 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,32,12.657,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 134 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,33,15.077,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 135 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,34,10.434,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 136 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,35,9.643,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 137 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,36,10.076,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 138 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,37,8.152,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 139 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,38,9.798,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 140 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,39,9.833,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 141 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,40,8.749,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 142 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,41,8.207,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 143 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,42,10.473,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 144 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,43,10.312,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 145 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,44,9.183,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 146 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,45,10.323,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 147 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,46,10.461,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 148 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,47,9.647,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 149 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,48,9.461,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 150 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,49,9.144,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 151 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,50,10.836,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 152 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,51,11.123,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 153 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,52,9.915,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 154 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,53,10.636,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 155 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,54,9.505,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 156 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,55,8.657,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 157 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,56,10.046,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 158 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,57,10.223,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 159 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,58,9.43,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 160 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,59,8.777,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 161 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,60,10.198,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 162 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,61,8.8,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 163 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,62,9.933,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 164 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,63,9.452,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 165 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,64,8.612,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 166 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,65,11.873,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 167 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,66,8.27,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 168 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,67,9.883,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 169 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,68,9.079,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 170 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,69,8.338,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 171 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,70,14.099,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 172 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,71,8.422,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 173 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,72,12.051,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 174 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,73,14.075,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 175 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,74,10.87,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 176 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,75,12.102,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 177 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,76,11.625,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 178 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,77,11.766,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 179 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,78,12.59,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 180 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,79,9.785,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 181 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,80,9.51,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 182 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,81,9.644,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 183 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,82,10.018,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 184 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,83,8.537,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 185 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,84,10.525,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 186 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,85,9.411,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 187 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,86,9.993,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 188 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,87,9.976,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 189 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,88,8.733,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 190 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,89,9.998,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 191 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,90,7.718,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 192 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,91,10.805,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 193 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,92,10.543,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 194 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,93,8.499,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 195 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,94,9.656,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 196 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,95,8.78,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 197 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,96,9.787,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 198 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,97,9.033,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 199 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,98,10.786,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 200 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,99,9.65,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 201 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,GPU_RTX5060,100,9.833,96.88%,93/96,W4A8_Tensor_Cores,CUDAExecutionProvider,Workstation_Machine_1
|
| 202 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,1,97.778,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 203 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,2,74.217,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 204 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,3,68.665,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 205 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,4,72.384,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 206 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,5,68.876,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 207 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,6,80.623,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 208 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,7,81.046,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 209 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,8,72.903,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 210 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,9,58.107,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 211 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,10,64.503,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 212 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,11,88.216,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 213 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,12,62.214,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 214 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,13,94.059,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 215 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,14,114.597,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 216 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,15,76.698,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 217 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,16,67.607,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 218 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,17,70.142,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 219 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,18,79.649,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 220 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,19,73.438,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 221 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,20,66.516,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 222 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,21,68.818,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 223 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,22,75.077,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 224 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,23,134.861,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 225 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,24,101.314,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 226 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,25,103.649,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 227 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,26,51.961,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 228 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,27,55.081,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 229 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,28,73.367,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 230 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,29,83.108,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 231 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,30,62.612,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 232 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,31,64.673,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 233 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,32,60.194,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 234 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,33,63.116,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 235 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,34,59.039,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 236 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,35,57.419,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 237 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,36,58.416,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 238 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,37,55.556,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 239 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,38,63.82,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 240 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,39,60.419,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 241 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,40,59.359,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 242 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,41,55.659,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 243 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,42,55.611,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 244 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,43,52.499,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 245 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,44,54.545,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 246 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,45,53.617,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 247 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,46,55.083,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 248 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,47,55.013,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 249 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,48,53.347,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 250 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,49,65.522,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 251 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,50,58.83,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 252 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,51,104.438,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 253 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,52,72.638,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 254 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,53,52.573,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 255 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,54,59.391,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 256 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,55,57.011,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 257 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,56,63.964,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 258 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,57,75.327,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 259 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,58,78.782,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 260 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,59,66.153,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 261 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,60,101.856,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 262 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,61,98.772,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 263 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,62,77.495,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 264 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,63,57.731,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 265 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,64,70.685,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 266 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,65,62.598,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 267 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,66,55.938,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 268 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,67,88.442,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 269 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,68,134.387,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 270 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,69,71.469,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 271 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,70,70.01,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 272 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,71,59.257,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 273 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,72,64.205,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 274 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,73,53.591,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 275 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,74,57.891,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 276 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,75,82.568,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 277 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,76,63.849,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 278 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,77,61.471,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 279 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,78,64.63,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 280 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,79,57.432,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 281 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,80,70.97,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 282 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,81,63.204,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 283 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,82,61.099,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 284 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,83,83.829,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 285 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,84,81.501,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 286 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,85,78.102,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 287 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,86,80.014,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 288 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,87,63.365,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 289 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,88,57.514,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 290 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,89,60.108,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 291 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,90,55.267,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 292 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,91,54.911,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 293 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,92,56.943,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 294 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,93,52.347,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 295 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,94,51.776,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 296 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,95,51.83,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 297 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,96,60.852,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 298 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,97,85.815,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 299 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,98,118.967,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 300 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,99,67.718,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 301 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,CPU_AVX2,100,93.184,96.88%,93/96,INT8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 302 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,1,87.543,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 303 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,2,70.708,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 304 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,3,61.432,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 305 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,4,81.652,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 306 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,5,98.374,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 307 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,6,100.273,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 308 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,7,83.261,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 309 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,8,321.49,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 310 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,9,152.347,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 311 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,10,154.87,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 312 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,11,91.849,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 313 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,12,66.189,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 314 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,13,70.938,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 315 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,14,89.79,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 316 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,15,93.031,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 317 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,16,73.366,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 318 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,17,75.824,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 319 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,18,59.679,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 320 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,19,79.951,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 321 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,20,102.009,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 322 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,21,91.948,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 323 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,22,75.38,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 324 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,23,88.679,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 325 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,24,70.995,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 326 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,25,57.864,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 327 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,26,50.507,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 328 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,27,54.296,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 329 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,28,49.024,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 330 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,29,53.077,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 331 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,30,66.439,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 332 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,31,61.131,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 333 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,32,57.162,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 334 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,33,48.936,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 335 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,34,55.361,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 336 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,35,54.333,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 337 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,36,64.978,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 338 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,37,68.679,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 339 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,38,53.497,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 340 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,39,51.915,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 341 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,40,60.72,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 342 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,41,58.074,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 343 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,42,53.72,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 344 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,43,61.374,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 345 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,44,64.046,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 346 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,45,74.841,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 347 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,46,68.54,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 348 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,47,85.679,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 349 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,48,68.93,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 350 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,49,57.075,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 351 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,50,84.577,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 352 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,51,52.447,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 353 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,52,49.56,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 354 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,53,49.872,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 355 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,54,58.543,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 356 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,55,59.665,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 357 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,56,63.768,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 358 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,57,59.554,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 359 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,58,74.883,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 360 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,59,59.582,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 361 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,60,52.826,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 362 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,61,56.41,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 363 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,62,54.003,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 364 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,63,57.029,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 365 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,64,52.58,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 366 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,65,59.05,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 367 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,66,64.3,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 368 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,67,72.594,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 369 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,68,52.134,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 370 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,69,48.771,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 371 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,70,56.53,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 372 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,71,49.388,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 373 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,72,49.175,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 374 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,73,55.003,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 375 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,74,56.151,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 376 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,75,58.731,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 377 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,76,65.553,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 378 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,77,58.239,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 379 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,78,54.75,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 380 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,79,54.516,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 381 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,80,48.666,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 382 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,81,75.027,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 383 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,82,58.326,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 384 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,83,51.854,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 385 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,84,82.06,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 386 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,85,51.11,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 387 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,86,50.043,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 388 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,87,62.633,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 389 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,88,56.093,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 390 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,89,59.947,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 391 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,90,50.145,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 392 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,91,47.576,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 393 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,92,51.029,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 394 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,93,55.321,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 395 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,94,54.804,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 396 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,95,49.951,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 397 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,96,52.849,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 398 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,97,49.13,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 399 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,98,47.265,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 400 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,99,45.79,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
| 401 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,CPU_AVX2,100,52.221,96.88%,93/96,W4A8_AVX2,CPUExecutionProvider,Workstation_Machine_1
|
results/ScaleDetection/machine_1/scaledetection_machine_1_comparison.png
ADDED
|
results/ScaleDetection/machine_2/benchmark_machine_2_scale_comparative.csv
ADDED
|
@@ -0,0 +1,601 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Model,Task,Format,Backend,Inference_ID,Latency_ms,Accuracy
|
| 2 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,1,17.462,77.14%
|
| 3 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,2,18.806,77.14%
|
| 4 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,3,17.976,77.14%
|
| 5 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,4,18.406,77.14%
|
| 6 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,5,18.462,77.14%
|
| 7 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,6,17.809,77.14%
|
| 8 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,7,19.198,77.14%
|
| 9 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,8,18.05,77.14%
|
| 10 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,9,19.164,77.14%
|
| 11 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,10,18.885,77.14%
|
| 12 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,11,19.487,77.14%
|
| 13 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,12,19.105,77.14%
|
| 14 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,13,18.5,77.14%
|
| 15 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,14,18.983,77.14%
|
| 16 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,15,17.915,77.14%
|
| 17 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,16,18.388,77.14%
|
| 18 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,17,19.246,77.14%
|
| 19 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,18,18.362,77.14%
|
| 20 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,19,19.084,77.14%
|
| 21 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,20,18.724,77.14%
|
| 22 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,21,18.788,77.14%
|
| 23 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,22,17.575,77.14%
|
| 24 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,23,18.045,77.14%
|
| 25 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,24,17.662,77.14%
|
| 26 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,25,18.25,77.14%
|
| 27 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,26,18.995,77.14%
|
| 28 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,27,18.498,77.14%
|
| 29 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,28,18.159,77.14%
|
| 30 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,29,19.619,77.14%
|
| 31 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,30,19.247,77.14%
|
| 32 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,31,19.688,77.14%
|
| 33 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,32,19.702,77.14%
|
| 34 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,33,19.686,77.14%
|
| 35 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,34,18.664,77.14%
|
| 36 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,35,17.39,77.14%
|
| 37 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,36,18.265,77.14%
|
| 38 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,37,19.299,77.14%
|
| 39 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,38,18.322,77.14%
|
| 40 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,39,18.996,77.14%
|
| 41 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,40,18.511,77.14%
|
| 42 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,41,19.608,77.14%
|
| 43 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,42,18.217,77.14%
|
| 44 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,43,18.363,77.14%
|
| 45 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,44,18.702,77.14%
|
| 46 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,45,19.028,77.14%
|
| 47 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,46,17.846,77.14%
|
| 48 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,47,18.286,77.14%
|
| 49 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,48,19.49,77.14%
|
| 50 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,49,18.636,77.14%
|
| 51 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,50,18.372,77.14%
|
| 52 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,51,18.399,77.14%
|
| 53 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,52,19.535,77.14%
|
| 54 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,53,19.115,77.14%
|
| 55 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,54,18.149,77.14%
|
| 56 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,55,18.482,77.14%
|
| 57 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,56,18.187,77.14%
|
| 58 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,57,18.707,77.14%
|
| 59 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,58,17.475,77.14%
|
| 60 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,59,17.786,77.14%
|
| 61 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,60,19.037,77.14%
|
| 62 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,61,19.101,77.14%
|
| 63 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,62,18.239,77.14%
|
| 64 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,63,18.337,77.14%
|
| 65 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,64,17.752,77.14%
|
| 66 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,65,19.271,77.14%
|
| 67 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,66,17.908,77.14%
|
| 68 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,67,19.173,77.14%
|
| 69 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,68,19.315,77.14%
|
| 70 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,69,17.875,77.14%
|
| 71 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,70,17.262,77.14%
|
| 72 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,71,18.933,77.14%
|
| 73 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,72,19.598,77.14%
|
| 74 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,73,18.926,77.14%
|
| 75 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,74,17.553,77.14%
|
| 76 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,75,19.148,77.14%
|
| 77 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,76,18.047,77.14%
|
| 78 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,77,18.344,77.14%
|
| 79 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,78,18.316,77.14%
|
| 80 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,79,18.023,77.14%
|
| 81 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,80,18.124,77.14%
|
| 82 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,81,18.862,77.14%
|
| 83 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,82,18.304,77.14%
|
| 84 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,83,18.54,77.14%
|
| 85 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,84,18.163,77.14%
|
| 86 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,85,17.575,77.14%
|
| 87 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,86,17.673,77.14%
|
| 88 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,87,19.102,77.14%
|
| 89 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,88,18.735,77.14%
|
| 90 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,89,19.335,77.14%
|
| 91 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,90,17.783,77.14%
|
| 92 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,91,17.918,77.14%
|
| 93 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,92,17.613,77.14%
|
| 94 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,93,18.937,77.14%
|
| 95 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,94,18.361,77.14%
|
| 96 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,95,18.53,77.14%
|
| 97 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,96,17.4,77.14%
|
| 98 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,97,18.192,77.14%
|
| 99 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,98,18.531,77.14%
|
| 100 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,99,17.881,77.14%
|
| 101 |
+
MobileNetV4-Small,Scale_Metrology,INT8,NPU,100,18.29,77.14%
|
| 102 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,1,112.766,77.14%
|
| 103 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,2,110.715,77.14%
|
| 104 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,3,114.172,77.14%
|
| 105 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,4,107.35,77.14%
|
| 106 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,5,110.198,77.14%
|
| 107 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,6,115.53,77.14%
|
| 108 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,7,109.945,77.14%
|
| 109 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,8,118.829,77.14%
|
| 110 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,9,125.501,77.14%
|
| 111 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,10,114.115,77.14%
|
| 112 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,11,115.489,77.14%
|
| 113 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,12,119.698,77.14%
|
| 114 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,13,116.49,77.14%
|
| 115 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,14,119.236,77.14%
|
| 116 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,15,116.788,77.14%
|
| 117 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,16,109.967,77.14%
|
| 118 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,17,117.863,77.14%
|
| 119 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,18,114.572,77.14%
|
| 120 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,19,110.621,77.14%
|
| 121 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,20,108.453,77.14%
|
| 122 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,21,109.535,77.14%
|
| 123 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,22,118.796,77.14%
|
| 124 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,23,115.714,77.14%
|
| 125 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,24,117.51,77.14%
|
| 126 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,25,114.68,77.14%
|
| 127 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,26,111.351,77.14%
|
| 128 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,27,112.225,77.14%
|
| 129 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,28,120.22,77.14%
|
| 130 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,29,118.74,77.14%
|
| 131 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,30,122.686,77.14%
|
| 132 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,31,112.422,77.14%
|
| 133 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,32,114.616,77.14%
|
| 134 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,33,118.727,77.14%
|
| 135 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,34,116.909,77.14%
|
| 136 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,35,122.887,77.14%
|
| 137 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,36,112.515,77.14%
|
| 138 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,37,112.747,77.14%
|
| 139 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,38,111.509,77.14%
|
| 140 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,39,122.077,77.14%
|
| 141 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,40,112.913,77.14%
|
| 142 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,41,117.617,77.14%
|
| 143 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,42,107.779,77.14%
|
| 144 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,43,116.548,77.14%
|
| 145 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,44,119.003,77.14%
|
| 146 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,45,116.089,77.14%
|
| 147 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,46,112.211,77.14%
|
| 148 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,47,115.999,77.14%
|
| 149 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,48,113.799,77.14%
|
| 150 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,49,111.345,77.14%
|
| 151 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,50,107.736,77.14%
|
| 152 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,51,118.121,77.14%
|
| 153 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,52,112.725,77.14%
|
| 154 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,53,108.895,77.14%
|
| 155 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,54,118.46,77.14%
|
| 156 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,55,117.968,77.14%
|
| 157 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,56,115.264,77.14%
|
| 158 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,57,116.445,77.14%
|
| 159 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,58,114.213,77.14%
|
| 160 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,59,108.374,77.14%
|
| 161 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,60,125.108,77.14%
|
| 162 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,61,119.276,77.14%
|
| 163 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,62,108.402,77.14%
|
| 164 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,63,110.852,77.14%
|
| 165 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,64,118.523,77.14%
|
| 166 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,65,121.524,77.14%
|
| 167 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,66,111.374,77.14%
|
| 168 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,67,113.162,77.14%
|
| 169 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,68,119.874,77.14%
|
| 170 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,69,121.221,77.14%
|
| 171 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,70,116.75,77.14%
|
| 172 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,71,114.504,77.14%
|
| 173 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,72,108.877,77.14%
|
| 174 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,73,117.24,77.14%
|
| 175 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,74,115.274,77.14%
|
| 176 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,75,111.923,77.14%
|
| 177 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,76,114.265,77.14%
|
| 178 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,77,122.857,77.14%
|
| 179 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,78,121.078,77.14%
|
| 180 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,79,118.346,77.14%
|
| 181 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,80,109.271,77.14%
|
| 182 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,81,122.751,77.14%
|
| 183 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,82,111.252,77.14%
|
| 184 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,83,116.446,77.14%
|
| 185 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,84,110.34,77.14%
|
| 186 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,85,113.238,77.14%
|
| 187 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,86,119.492,77.14%
|
| 188 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,87,113.536,77.14%
|
| 189 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,88,114.367,77.14%
|
| 190 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,89,115.32,77.14%
|
| 191 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,90,112.506,77.14%
|
| 192 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,91,118.693,77.14%
|
| 193 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,92,114.728,77.14%
|
| 194 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,93,114.055,77.14%
|
| 195 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,94,112.658,77.14%
|
| 196 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,95,109.991,77.14%
|
| 197 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,96,115.203,77.14%
|
| 198 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,97,109.512,77.14%
|
| 199 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,98,107.685,77.14%
|
| 200 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,99,113.714,77.14%
|
| 201 |
+
MobileNetV4-Small,Scale_Metrology,INT8,GPU,100,112.837,77.14%
|
| 202 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,1,125.937,77.14%
|
| 203 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,2,118.164,77.14%
|
| 204 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,3,115.96,77.14%
|
| 205 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,4,115.305,77.14%
|
| 206 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,5,125.125,77.14%
|
| 207 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,6,112.656,77.14%
|
| 208 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,7,117.735,77.14%
|
| 209 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,8,116.588,77.14%
|
| 210 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,9,117.812,77.14%
|
| 211 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,10,114.47,77.14%
|
| 212 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,11,110.812,77.14%
|
| 213 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,12,112.062,77.14%
|
| 214 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,13,123.435,77.14%
|
| 215 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,14,124.781,77.14%
|
| 216 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,15,113.495,77.14%
|
| 217 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,16,118.133,77.14%
|
| 218 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,17,115.661,77.14%
|
| 219 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,18,122.756,77.14%
|
| 220 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,19,114.884,77.14%
|
| 221 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,20,122.384,77.14%
|
| 222 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,21,114.942,77.14%
|
| 223 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,22,115.349,77.14%
|
| 224 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,23,102.569,77.14%
|
| 225 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,24,118.32,77.14%
|
| 226 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,25,122.754,77.14%
|
| 227 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,26,109.193,77.14%
|
| 228 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,27,115.919,77.14%
|
| 229 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,28,117.081,77.14%
|
| 230 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,29,118.494,77.14%
|
| 231 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,30,127.703,77.14%
|
| 232 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,31,128.078,77.14%
|
| 233 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,32,112.865,77.14%
|
| 234 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,33,116.902,77.14%
|
| 235 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,34,119.888,77.14%
|
| 236 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,35,116.158,77.14%
|
| 237 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,36,109.833,77.14%
|
| 238 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,37,119.349,77.14%
|
| 239 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,38,110.848,77.14%
|
| 240 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,39,127.059,77.14%
|
| 241 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,40,121.094,77.14%
|
| 242 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,41,119.068,77.14%
|
| 243 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,42,114.525,77.14%
|
| 244 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,43,115.747,77.14%
|
| 245 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,44,110.928,77.14%
|
| 246 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,45,115.321,77.14%
|
| 247 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,46,121.044,77.14%
|
| 248 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,47,115.26,77.14%
|
| 249 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,48,115.882,77.14%
|
| 250 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,49,109.942,77.14%
|
| 251 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,50,113.805,77.14%
|
| 252 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,51,110.484,77.14%
|
| 253 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,52,116.139,77.14%
|
| 254 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,53,106.886,77.14%
|
| 255 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,54,108.269,77.14%
|
| 256 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,55,106.356,77.14%
|
| 257 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,56,117.098,77.14%
|
| 258 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,57,120.509,77.14%
|
| 259 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,58,117.198,77.14%
|
| 260 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,59,122.249,77.14%
|
| 261 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,60,122.21,77.14%
|
| 262 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,61,113.357,77.14%
|
| 263 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,62,125.801,77.14%
|
| 264 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,63,116.58,77.14%
|
| 265 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,64,114.665,77.14%
|
| 266 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,65,111.306,77.14%
|
| 267 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,66,112.446,77.14%
|
| 268 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,67,114.385,77.14%
|
| 269 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,68,113.495,77.14%
|
| 270 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,69,118.129,77.14%
|
| 271 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,70,111.637,77.14%
|
| 272 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,71,117.011,77.14%
|
| 273 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,72,112.673,77.14%
|
| 274 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,73,120.466,77.14%
|
| 275 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,74,119.875,77.14%
|
| 276 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,75,119.538,77.14%
|
| 277 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,76,129.313,77.14%
|
| 278 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,77,117.236,77.14%
|
| 279 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,78,117.675,77.14%
|
| 280 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,79,117.991,77.14%
|
| 281 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,80,116.478,77.14%
|
| 282 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,81,111.799,77.14%
|
| 283 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,82,117.454,77.14%
|
| 284 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,83,118.611,77.14%
|
| 285 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,84,111.644,77.14%
|
| 286 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,85,116.053,77.14%
|
| 287 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,86,107.791,77.14%
|
| 288 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,87,116.368,77.14%
|
| 289 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,88,116.46,77.14%
|
| 290 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,89,115.504,77.14%
|
| 291 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,90,116.757,77.14%
|
| 292 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,91,110.059,77.14%
|
| 293 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,92,115.613,77.14%
|
| 294 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,93,118.427,77.14%
|
| 295 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,94,114.257,77.14%
|
| 296 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,95,114.567,77.14%
|
| 297 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,96,111.576,77.14%
|
| 298 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,97,110.568,77.14%
|
| 299 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,98,111.685,77.14%
|
| 300 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,99,116.695,77.14%
|
| 301 |
+
MobileNetV4-Small,Scale_Metrology,INT8,CPU,100,123.07,77.14%
|
| 302 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,1,12.111,77.14%
|
| 303 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,2,11.119,77.14%
|
| 304 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,3,12.022,77.14%
|
| 305 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,4,11.887,77.14%
|
| 306 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,5,12.226,77.14%
|
| 307 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,6,11.663,77.14%
|
| 308 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,7,12.048,77.14%
|
| 309 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,8,11.751,77.14%
|
| 310 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,9,12.512,77.14%
|
| 311 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,10,11.946,77.14%
|
| 312 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,11,12.122,77.14%
|
| 313 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,12,11.352,77.14%
|
| 314 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,13,11.986,77.14%
|
| 315 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,14,11.622,77.14%
|
| 316 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,15,11.947,77.14%
|
| 317 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,16,12.19,77.14%
|
| 318 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,17,11.443,77.14%
|
| 319 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,18,12.442,77.14%
|
| 320 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,19,11.921,77.14%
|
| 321 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,20,12.504,77.14%
|
| 322 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,21,12.066,77.14%
|
| 323 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,22,11.69,77.14%
|
| 324 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,23,12.651,77.14%
|
| 325 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,24,12.561,77.14%
|
| 326 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,25,12.231,77.14%
|
| 327 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,26,10.978,77.14%
|
| 328 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,27,12.484,77.14%
|
| 329 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,28,11.754,77.14%
|
| 330 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,29,11.664,77.14%
|
| 331 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,30,11.783,77.14%
|
| 332 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,31,12.766,77.14%
|
| 333 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,32,11.187,77.14%
|
| 334 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,33,12.413,77.14%
|
| 335 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,34,11.016,77.14%
|
| 336 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,35,12.86,77.14%
|
| 337 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,36,12.378,77.14%
|
| 338 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,37,11.721,77.14%
|
| 339 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,38,11.486,77.14%
|
| 340 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,39,12.262,77.14%
|
| 341 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,40,11.425,77.14%
|
| 342 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,41,12.228,77.14%
|
| 343 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,42,11.771,77.14%
|
| 344 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,43,12.435,77.14%
|
| 345 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,44,11.776,77.14%
|
| 346 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,45,12.189,77.14%
|
| 347 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,46,11.928,77.14%
|
| 348 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,47,12.469,77.14%
|
| 349 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,48,11.637,77.14%
|
| 350 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,49,13.12,77.14%
|
| 351 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,50,12.749,77.14%
|
| 352 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,51,12.467,77.14%
|
| 353 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,52,11.135,77.14%
|
| 354 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,53,12.466,77.14%
|
| 355 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,54,11.952,77.14%
|
| 356 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,55,11.472,77.14%
|
| 357 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,56,11.947,77.14%
|
| 358 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,57,11.825,77.14%
|
| 359 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,58,11.761,77.14%
|
| 360 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,59,12.226,77.14%
|
| 361 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,60,12.268,77.14%
|
| 362 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,61,12.38,77.14%
|
| 363 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,62,11.936,77.14%
|
| 364 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,63,11.792,77.14%
|
| 365 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,64,11.941,77.14%
|
| 366 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,65,11.384,77.14%
|
| 367 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,66,11.486,77.14%
|
| 368 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,67,13.389,77.14%
|
| 369 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,68,12.306,77.14%
|
| 370 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,69,11.222,77.14%
|
| 371 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,70,11.74,77.14%
|
| 372 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,71,12.022,77.14%
|
| 373 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,72,11.932,77.14%
|
| 374 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,73,10.987,77.14%
|
| 375 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,74,11.446,77.14%
|
| 376 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,75,12.828,77.14%
|
| 377 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,76,11.548,77.14%
|
| 378 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,77,11.812,77.14%
|
| 379 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,78,11.692,77.14%
|
| 380 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,79,11.968,77.14%
|
| 381 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,80,11.72,77.14%
|
| 382 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,81,11.762,77.14%
|
| 383 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,82,11.132,77.14%
|
| 384 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,83,12.041,77.14%
|
| 385 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,84,12.544,77.14%
|
| 386 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,85,12.351,77.14%
|
| 387 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,86,12.191,77.14%
|
| 388 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,87,12.303,77.14%
|
| 389 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,88,11.586,77.14%
|
| 390 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,89,12.432,77.14%
|
| 391 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,90,11.552,77.14%
|
| 392 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,91,11.316,77.14%
|
| 393 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,92,13.225,77.14%
|
| 394 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,93,12.31,77.14%
|
| 395 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,94,12.37,77.14%
|
| 396 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,95,11.382,77.14%
|
| 397 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,96,11.762,77.14%
|
| 398 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,97,12.01,77.14%
|
| 399 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,98,11.72,77.14%
|
| 400 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,99,12.632,77.14%
|
| 401 |
+
MobileNetV4-Small,Scale_Metrology,INT4,NPU,100,12.24,77.14%
|
| 402 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,1,25.323,77.14%
|
| 403 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,2,24.263,77.14%
|
| 404 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,3,27.356,77.14%
|
| 405 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,4,26.367,77.14%
|
| 406 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,5,25.953,77.14%
|
| 407 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,6,25.174,77.14%
|
| 408 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,7,24.209,77.14%
|
| 409 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,8,25.437,77.14%
|
| 410 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,9,24.722,77.14%
|
| 411 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,10,24.896,77.14%
|
| 412 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,11,25.265,77.14%
|
| 413 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,12,26.005,77.14%
|
| 414 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,13,25.494,77.14%
|
| 415 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,14,25.175,77.14%
|
| 416 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,15,24.833,77.14%
|
| 417 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,16,25.178,77.14%
|
| 418 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,17,27.5,77.14%
|
| 419 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,18,26.691,77.14%
|
| 420 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,19,24.123,77.14%
|
| 421 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,20,25.43,77.14%
|
| 422 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,21,24.326,77.14%
|
| 423 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,22,25.959,77.14%
|
| 424 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,23,25.583,77.14%
|
| 425 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,24,25.157,77.14%
|
| 426 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,25,25.186,77.14%
|
| 427 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,26,27.13,77.14%
|
| 428 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,27,23.71,77.14%
|
| 429 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,28,26.746,77.14%
|
| 430 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,29,24.876,77.14%
|
| 431 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,30,25.436,77.14%
|
| 432 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,31,26.236,77.14%
|
| 433 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,32,24.56,77.14%
|
| 434 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,33,24.168,77.14%
|
| 435 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,34,25.739,77.14%
|
| 436 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,35,26.334,77.14%
|
| 437 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,36,25.895,77.14%
|
| 438 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,37,25.151,77.14%
|
| 439 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,38,25.47,77.14%
|
| 440 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,39,24.495,77.14%
|
| 441 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,40,26.822,77.14%
|
| 442 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,41,26.677,77.14%
|
| 443 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,42,24.598,77.14%
|
| 444 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,43,25.387,77.14%
|
| 445 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,44,25.686,77.14%
|
| 446 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,45,23.282,77.14%
|
| 447 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,46,25.972,77.14%
|
| 448 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,47,25.216,77.14%
|
| 449 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,48,25.583,77.14%
|
| 450 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,49,24.778,77.14%
|
| 451 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,50,24.718,77.14%
|
| 452 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,51,25.059,77.14%
|
| 453 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,52,25.856,77.14%
|
| 454 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,53,25.727,77.14%
|
| 455 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,54,25.184,77.14%
|
| 456 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,55,25.223,77.14%
|
| 457 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,56,25.048,77.14%
|
| 458 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,57,24.214,77.14%
|
| 459 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,58,25.572,77.14%
|
| 460 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,59,24.915,77.14%
|
| 461 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,60,26.129,77.14%
|
| 462 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,61,24.788,77.14%
|
| 463 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,62,24.504,77.14%
|
| 464 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,63,26.629,77.14%
|
| 465 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,64,26.074,77.14%
|
| 466 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,65,24.531,77.14%
|
| 467 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,66,27.605,77.14%
|
| 468 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,67,23.889,77.14%
|
| 469 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,68,24.755,77.14%
|
| 470 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,69,25.664,77.14%
|
| 471 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,70,24.431,77.14%
|
| 472 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,71,24.674,77.14%
|
| 473 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,72,24.228,77.14%
|
| 474 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,73,24.262,77.14%
|
| 475 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,74,25.244,77.14%
|
| 476 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,75,26.73,77.14%
|
| 477 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,76,25.43,77.14%
|
| 478 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,77,25.923,77.14%
|
| 479 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,78,25.282,77.14%
|
| 480 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,79,27.886,77.14%
|
| 481 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,80,25.245,77.14%
|
| 482 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,81,26.176,77.14%
|
| 483 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,82,26.137,77.14%
|
| 484 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,83,23.738,77.14%
|
| 485 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,84,23.483,77.14%
|
| 486 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,85,26.541,77.14%
|
| 487 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,86,23.651,77.14%
|
| 488 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,87,24.65,77.14%
|
| 489 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,88,25.4,77.14%
|
| 490 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,89,25.705,77.14%
|
| 491 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,90,25.199,77.14%
|
| 492 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,91,26.758,77.14%
|
| 493 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,92,25.053,77.14%
|
| 494 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,93,25.477,77.14%
|
| 495 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,94,26.389,77.14%
|
| 496 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,95,25.793,77.14%
|
| 497 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,96,25.962,77.14%
|
| 498 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,97,25.109,77.14%
|
| 499 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,98,25.666,77.14%
|
| 500 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,99,25.646,77.14%
|
| 501 |
+
MobileNetV4-Small,Scale_Metrology,INT4,GPU,100,25.248,77.14%
|
| 502 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,1,426.887,77.14%
|
| 503 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,2,406.563,77.14%
|
| 504 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,3,384.648,77.14%
|
| 505 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,4,404.127,77.14%
|
| 506 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,5,420.145,77.14%
|
| 507 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,6,398.929,77.14%
|
| 508 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,7,422.502,77.14%
|
| 509 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,8,396.795,77.14%
|
| 510 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,9,417.121,77.14%
|
| 511 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,10,425.53,77.14%
|
| 512 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,11,396.838,77.14%
|
| 513 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,12,420.411,77.14%
|
| 514 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,13,408.383,77.14%
|
| 515 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,14,415.261,77.14%
|
| 516 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,15,391.823,77.14%
|
| 517 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,16,419.731,77.14%
|
| 518 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,17,402.98,77.14%
|
| 519 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,18,433.753,77.14%
|
| 520 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,19,415.844,77.14%
|
| 521 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,20,400.225,77.14%
|
| 522 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,21,409.3,77.14%
|
| 523 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,22,417.584,77.14%
|
| 524 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,23,414.535,77.14%
|
| 525 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,24,426.271,77.14%
|
| 526 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,25,409.813,77.14%
|
| 527 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,26,430.863,77.14%
|
| 528 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,27,386.536,77.14%
|
| 529 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,28,427.436,77.14%
|
| 530 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,29,420.035,77.14%
|
| 531 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,30,438.849,77.14%
|
| 532 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,31,397.737,77.14%
|
| 533 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,32,404.593,77.14%
|
| 534 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,33,400.797,77.14%
|
| 535 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,34,410.33,77.14%
|
| 536 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,35,377.985,77.14%
|
| 537 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,36,382.212,77.14%
|
| 538 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,37,390.349,77.14%
|
| 539 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,38,421.201,77.14%
|
| 540 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,39,417.617,77.14%
|
| 541 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,40,401.892,77.14%
|
| 542 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,41,429.271,77.14%
|
| 543 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,42,421.33,77.14%
|
| 544 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,43,405.757,77.14%
|
| 545 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,44,421.266,77.14%
|
| 546 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,45,444.638,77.14%
|
| 547 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,46,429.971,77.14%
|
| 548 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,47,431.821,77.14%
|
| 549 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,48,418.075,77.14%
|
| 550 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,49,424.223,77.14%
|
| 551 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,50,407.33,77.14%
|
| 552 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,51,420.584,77.14%
|
| 553 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,52,397.718,77.14%
|
| 554 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,53,409.444,77.14%
|
| 555 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,54,404.209,77.14%
|
| 556 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,55,423.305,77.14%
|
| 557 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,56,432.58,77.14%
|
| 558 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,57,421.471,77.14%
|
| 559 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,58,426.853,77.14%
|
| 560 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,59,385.949,77.14%
|
| 561 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,60,402.492,77.14%
|
| 562 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,61,392.153,77.14%
|
| 563 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,62,430.251,77.14%
|
| 564 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,63,408.714,77.14%
|
| 565 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,64,427.724,77.14%
|
| 566 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,65,404.961,77.14%
|
| 567 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,66,385.75,77.14%
|
| 568 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,67,409.985,77.14%
|
| 569 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,68,415.178,77.14%
|
| 570 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,69,380.356,77.14%
|
| 571 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,70,420.505,77.14%
|
| 572 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,71,434.277,77.14%
|
| 573 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,72,406.725,77.14%
|
| 574 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,73,397.391,77.14%
|
| 575 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,74,421.805,77.14%
|
| 576 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,75,376.619,77.14%
|
| 577 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,76,437.391,77.14%
|
| 578 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,77,403.437,77.14%
|
| 579 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,78,403.861,77.14%
|
| 580 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,79,415.238,77.14%
|
| 581 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,80,406.411,77.14%
|
| 582 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,81,400.47,77.14%
|
| 583 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,82,403.825,77.14%
|
| 584 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,83,369.249,77.14%
|
| 585 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,84,421.262,77.14%
|
| 586 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,85,430.362,77.14%
|
| 587 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,86,438.239,77.14%
|
| 588 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,87,427.314,77.14%
|
| 589 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,88,411.04,77.14%
|
| 590 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,89,428.04,77.14%
|
| 591 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,90,431.344,77.14%
|
| 592 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,91,411.292,77.14%
|
| 593 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,92,405.968,77.14%
|
| 594 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,93,413.267,77.14%
|
| 595 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,94,376.136,77.14%
|
| 596 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,95,428.938,77.14%
|
| 597 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,96,431.374,77.14%
|
| 598 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,97,454.819,77.14%
|
| 599 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,98,415.865,77.14%
|
| 600 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,99,444.144,77.14%
|
| 601 |
+
MobileNetV4-Small,Scale_Metrology,INT4,CPU,100,412.909,77.14%
|
results/ScaleDetection/machine_2/benchmark_machine_2_scale_native_100_runs.csv
ADDED
|
@@ -0,0 +1,201 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Model,Task,Format,Backend,Inference_ID,Latency_ms,Top1_Accuracy,Top3_Accuracy,Hardware_Environment,Precision_Type,Execution_Status
|
| 2 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,1,510.274,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 3 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,2,513.075,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 4 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,3,520.126,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 5 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,4,590.644,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 6 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,5,589.83,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 7 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,6,563.352,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 8 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,7,596.937,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 9 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,8,555.194,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 10 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,9,541.438,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 11 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,10,558.363,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 12 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,11,548.104,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 13 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,12,564.473,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 14 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,13,519.354,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 15 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,14,536.956,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 16 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,15,484.168,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 17 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,16,579.975,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 18 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,17,575.403,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 19 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,18,545.676,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 20 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,19,515.435,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 21 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,20,485.331,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 22 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,21,563.424,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 23 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,22,552.306,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 24 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,23,540.529,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 25 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,24,610.88,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 26 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,25,546.848,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 27 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,26,555.461,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 28 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,27,598.453,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 29 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,28,532.724,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 30 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,29,531.019,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 31 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,30,545.718,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 32 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,31,552.18,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 33 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,32,522.197,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 34 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,33,550.949,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 35 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,34,561.429,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 36 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,35,528.455,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 37 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,36,545.984,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 38 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,37,589.848,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 39 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,38,488.26,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 40 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,39,542.809,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 41 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,40,509.216,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 42 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,41,505.569,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 43 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,42,492.731,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 44 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,43,514.608,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 45 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,44,520.645,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 46 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,45,591.667,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 47 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,46,530.075,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 48 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,47,519.549,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 49 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,48,530.147,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 50 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,49,523.943,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 51 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,50,500.048,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 52 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,51,560.719,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 53 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,52,556.574,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 54 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,53,512.816,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 55 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,54,546.413,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 56 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,55,531.283,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 57 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,56,479.613,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 58 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,57,572.401,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 59 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,58,504.786,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 60 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,59,477.553,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 61 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,60,455.646,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 62 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,61,451.16,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 63 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,62,499.761,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 64 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,63,516.042,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 65 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,64,506.93,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 66 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,65,533.549,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 67 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,66,536.725,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 68 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,67,544.305,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 69 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,68,598.06,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 70 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,69,544.532,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 71 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,70,545.381,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 72 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,71,506.901,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 73 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,72,561.972,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 74 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,73,526.449,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 75 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,74,482.183,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 76 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,75,571.515,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 77 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,76,496.486,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 78 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,77,514.317,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 79 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,78,556.753,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 80 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,79,471.193,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 81 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,80,530.765,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 82 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,81,482.295,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 83 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,82,494.773,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 84 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,83,503.396,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 85 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,84,614.182,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 86 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,85,522.446,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 87 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,86,511.163,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 88 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,87,577.811,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 89 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,88,563.605,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 90 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,89,545.007,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 91 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,90,516.781,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 92 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,91,542.969,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 93 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,92,489.736,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 94 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,93,539.646,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 95 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,94,545.631,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 96 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,95,512.1,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 97 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,96,592.931,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 98 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,97,620.58,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 99 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,98,584.749,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 100 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,99,487.87,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 101 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,W4A8,NPU,100,521.275,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W4A8_INT4_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 102 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,1,581.177,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 103 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,2,585.914,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 104 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,3,561.457,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 105 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,4,567.691,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 106 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,5,543.161,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 107 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,6,495.772,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 108 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,7,533.987,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 109 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,8,575.789,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 110 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,9,564.58,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 111 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,10,589.839,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 112 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,11,534.431,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 113 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,12,508.544,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 114 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,13,571.291,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 115 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,14,555.137,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 116 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,15,548.26,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 117 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,16,444.907,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 118 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,17,512.421,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 119 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,18,498.937,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 120 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,19,467.732,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 121 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,20,560.216,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 122 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,21,519.611,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 123 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,22,474.323,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 124 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,23,496.088,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 125 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,24,494.694,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 126 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,25,544.288,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 127 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,26,576.591,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 128 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,27,567.389,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 129 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,28,599.987,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 130 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,29,513.008,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 131 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,30,535.946,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 132 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,31,576.774,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 133 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,32,498.758,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 134 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,33,517.308,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 135 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,34,528.855,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 136 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,35,497.522,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 137 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,36,570.555,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 138 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,37,505.034,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 139 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,38,511.063,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 140 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,39,517.317,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 141 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,40,513.989,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 142 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,41,533.668,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 143 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,42,551.011,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 144 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,43,526.13,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 145 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,44,601.135,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 146 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,45,543.499,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 147 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,46,573.407,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 148 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,47,535.588,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 149 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,48,501.28,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 150 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,49,542.871,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 151 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,50,528.437,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 152 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,51,519.676,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 153 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,52,512.044,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 154 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,53,495.993,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 155 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,54,542.037,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 156 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,55,490.439,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 157 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,56,542.294,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 158 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,57,420.076,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 159 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,58,483.992,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 160 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,59,502.144,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 161 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,60,490.778,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 162 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,61,505.025,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 163 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,62,482.636,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 164 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,63,506.058,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 165 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,64,488.943,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 166 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,65,525.547,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 167 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,66,496.791,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 168 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,67,559.99,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 169 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,68,564.7,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 170 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,69,521.566,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 171 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,70,523.719,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 172 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,71,548.495,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 173 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,72,578.403,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 174 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,73,521.457,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 175 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,74,486.682,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 176 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,75,500.885,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 177 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,76,505.702,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 178 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,77,594.065,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 179 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,78,558.057,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 180 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,79,524.957,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 181 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,80,522.796,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 182 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,81,580.542,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 183 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,82,620.334,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 184 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,83,574.899,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 185 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,84,541.432,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 186 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,85,526.794,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 187 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,86,548.352,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 188 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,87,571.359,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 189 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,88,483.024,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 190 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,89,491.667,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 191 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,90,544.476,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 192 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,91,547.812,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 193 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,92,532.97,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 194 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,93,637.979,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 195 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,94,531.957,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 196 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,95,533.734,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 197 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,96,516.284,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 198 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,97,585.508,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 199 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,98,529.255,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 200 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,99,532.842,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
| 201 |
+
ScaleDetection_MobileNetV4,Scale_Metrology,INT8,NPU,100,496.644,96.88%,96.88%,Qualcomm_Snapdragon_X_Copilot_PC_ARM64_SP11,W8A8_INT8_HTP,Native_Hexagon_HTP_Hardware_Accelerated_Empirical_Eval_96
|
results/ScaleDetection/machine_2/scale_metrology_detection_profiling.png
ADDED
|
results/ScaleDetection/machine_2/scaledetection_machine_2_native_stability.png
ADDED
|
scripts/benchmark_npu_onnx.py
ADDED
|
@@ -0,0 +1,189 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import glob
|
| 3 |
+
import json
|
| 4 |
+
import time
|
| 5 |
+
import re
|
| 6 |
+
import argparse
|
| 7 |
+
import numpy as np
|
| 8 |
+
import onnxruntime as ort
|
| 9 |
+
from PIL import Image
|
| 10 |
+
|
| 11 |
+
def clean(s: str) -> str:
|
| 12 |
+
return re.sub(r'[^a-z0-9]', '', s.lower())
|
| 13 |
+
|
| 14 |
+
def resolve_label(label_raw: str, c104_map: dict, synonyms: dict):
|
| 15 |
+
ctl = clean(label_raw)
|
| 16 |
+
if ctl in c104_map:
|
| 17 |
+
return c104_map[ctl]
|
| 18 |
+
if ctl in synonyms and synonyms[ctl] in c104_map:
|
| 19 |
+
return c104_map[synonyms[ctl]]
|
| 20 |
+
for k, v in c104_map.items():
|
| 21 |
+
if ctl in k or k in ctl:
|
| 22 |
+
return v
|
| 23 |
+
return None
|
| 24 |
+
|
| 25 |
+
def main():
|
| 26 |
+
parser = argparse.ArgumentParser(description="Benchmark ONNX models on Snapdragon Hexagon NPU via QNN Execution Provider.")
|
| 27 |
+
parser.add_argument("--classes", type=str, default="classes_104_en.json", help="Path to classes JSON list.")
|
| 28 |
+
parser.add_argument("--test_dir", type=str, default="test_images", help="Directory containing test images.")
|
| 29 |
+
parser.add_argument("--models_dir", type=str, default="models", help="Directory containing .onnx models.")
|
| 30 |
+
parser.add_argument("--backend_path", type=str, default="QnnHtp.dll", help="Path or library name for QNN HTP backend.")
|
| 31 |
+
parser.add_argument("--output_json", type=str, default="benchmark_results_npu.json", help="File path to store output metrics.")
|
| 32 |
+
args = parser.parse_args()
|
| 33 |
+
|
| 34 |
+
if not os.path.exists(args.classes):
|
| 35 |
+
print(f"[!] Warning: Classes file '{args.classes}' not found. Please provide valid path.")
|
| 36 |
+
return
|
| 37 |
+
|
| 38 |
+
with open(args.classes, "r", encoding="utf-8") as f:
|
| 39 |
+
classes = json.load(f)
|
| 40 |
+
|
| 41 |
+
c104_map = {clean(x): idx for idx, x in enumerate(classes)}
|
| 42 |
+
|
| 43 |
+
synonyms = {
|
| 44 |
+
'coquine': clean('Coquina (Shell limestone)'),
|
| 45 |
+
'dolomiticlimestone': clean('Dolomite limestone'),
|
| 46 |
+
'shalemudstone': clean('Shales'),
|
| 47 |
+
'stibnite': clean('Stibine'),
|
| 48 |
+
'breccia': clean('Breach'),
|
| 49 |
+
'chert': clean('Chert (Silexite)'),
|
| 50 |
+
'hornfels': clean('Hornfels (Cornea)'),
|
| 51 |
+
'novaculite': clean('Novaculitis'),
|
| 52 |
+
'pillow': clean('Pillow lava (Pillow)'),
|
| 53 |
+
'silver': clean('Money'),
|
| 54 |
+
'tufa': clean('Tuff (Liestone)'),
|
| 55 |
+
'tuff': clean('Tuff (Volcanic)'),
|
| 56 |
+
'schists': clean('Schist (Limonite)')
|
| 57 |
+
}
|
| 58 |
+
|
| 59 |
+
img_files = sorted(glob.glob(os.path.join(args.test_dir, "*.jpg"))) + sorted(glob.glob(os.path.join(args.test_dir, "*.png")))
|
| 60 |
+
print(f"Loaded {len(img_files)} test images for evaluation.")
|
| 61 |
+
|
| 62 |
+
qnn_options = {
|
| 63 |
+
"backend_path": args.backend_path,
|
| 64 |
+
"htp_performance_mode": "burst",
|
| 65 |
+
"htp_graph_finalization_optimization_mode": "3"
|
| 66 |
+
}
|
| 67 |
+
|
| 68 |
+
models_to_bench = [
|
| 69 |
+
("MobileNetV4 INT8", os.path.join(args.models_dir, "rock_model_int8_MobileNetV4.onnx"), "INT8"),
|
| 70 |
+
("MobileNetV4 W4A8", os.path.join(args.models_dir, "rock_model_w4a8_MobileNetV4.onnx"), "W4A8")
|
| 71 |
+
]
|
| 72 |
+
|
| 73 |
+
results_summary = {}
|
| 74 |
+
|
| 75 |
+
for label, model_path, fmt in models_to_bench:
|
| 76 |
+
if not os.path.exists(model_path):
|
| 77 |
+
print(f"[-] Skipping {label}: '{model_path}' not found.")
|
| 78 |
+
continue
|
| 79 |
+
|
| 80 |
+
print(f"\n=======================================================")
|
| 81 |
+
print(f"STARTING EVALUATION: {label} on Hexagon NPU")
|
| 82 |
+
print(f"=======================================================")
|
| 83 |
+
|
| 84 |
+
sess_opts = ort.SessionOptions()
|
| 85 |
+
sess_opts.graph_optimization_level = ort.GraphOptimizationLevel.ORT_ENABLE_ALL
|
| 86 |
+
|
| 87 |
+
session = ort.InferenceSession(
|
| 88 |
+
model_path,
|
| 89 |
+
sess_options=sess_opts,
|
| 90 |
+
providers=["QNNExecutionProvider", "CPUExecutionProvider"],
|
| 91 |
+
provider_options=[qnn_options, {}]
|
| 92 |
+
)
|
| 93 |
+
print(f"Active Provider: {session.get_providers()[0]}")
|
| 94 |
+
|
| 95 |
+
inp_name = session.get_inputs()[0].name
|
| 96 |
+
actual_shape = session.get_inputs()[0].shape
|
| 97 |
+
inp_type = session.get_inputs()[0].type
|
| 98 |
+
out_name = session.get_outputs()[0].name
|
| 99 |
+
print(f"Input Name: {inp_name} | Type: {inp_type} | Shape: {actual_shape}")
|
| 100 |
+
|
| 101 |
+
is_nchw = (len(actual_shape) == 4 and actual_shape[1] == 3)
|
| 102 |
+
if is_nchw:
|
| 103 |
+
H, W = actual_shape[2], actual_shape[3]
|
| 104 |
+
else:
|
| 105 |
+
H, W = actual_shape[1], actual_shape[2]
|
| 106 |
+
|
| 107 |
+
print(f"Target resolution: {H}x{W} (Layout: {'NCHW' if is_nchw else 'NHWC'})")
|
| 108 |
+
|
| 109 |
+
top1_correct = 0
|
| 110 |
+
top3_correct = 0
|
| 111 |
+
total_evaluated = 0
|
| 112 |
+
latencies = []
|
| 113 |
+
|
| 114 |
+
# Warmup passes
|
| 115 |
+
dummy = np.random.randint(-128, 127, size=[d if isinstance(d, int) and d > 0 else 1 for d in actual_shape], dtype=np.int8)
|
| 116 |
+
for _ in range(5):
|
| 117 |
+
_ = session.run([out_name], {inp_name: dummy})
|
| 118 |
+
|
| 119 |
+
for idx, img_path in enumerate(img_files):
|
| 120 |
+
base = os.path.splitext(os.path.basename(img_path))[0]
|
| 121 |
+
parts = base.split("_", 2)
|
| 122 |
+
if len(parts) < 3:
|
| 123 |
+
continue
|
| 124 |
+
true_label_raw = parts[2]
|
| 125 |
+
true_idx = resolve_label(true_label_raw, c104_map, synonyms)
|
| 126 |
+
if true_idx is None:
|
| 127 |
+
continue
|
| 128 |
+
|
| 129 |
+
img = Image.open(img_path).convert("RGB").resize((W, H))
|
| 130 |
+
arr = np.array(img, dtype=np.float32)
|
| 131 |
+
norm_arr = (arr - 128.0) / 127.5
|
| 132 |
+
q_inp = np.clip(np.round(norm_arr / 0.008305505849421024) + 127, -128, 127).astype(np.int8)
|
| 133 |
+
|
| 134 |
+
if is_nchw:
|
| 135 |
+
q_inp = np.transpose(q_inp, (2, 0, 1))
|
| 136 |
+
q_inp = np.expand_dims(q_inp, axis=0)
|
| 137 |
+
|
| 138 |
+
t0 = time.perf_counter()
|
| 139 |
+
res = session.run([out_name], {inp_name: q_inp})[0]
|
| 140 |
+
t1 = time.perf_counter()
|
| 141 |
+
lat_ms = (t1 - t0) * 1000.0
|
| 142 |
+
latencies.append(lat_ms)
|
| 143 |
+
|
| 144 |
+
logits = res[0]
|
| 145 |
+
top_indices = np.argsort(logits)[::-1]
|
| 146 |
+
|
| 147 |
+
if top_indices[0] == true_idx:
|
| 148 |
+
top1_correct += 1
|
| 149 |
+
if true_idx in top_indices[:3]:
|
| 150 |
+
top3_correct += 1
|
| 151 |
+
total_evaluated += 1
|
| 152 |
+
|
| 153 |
+
top1_acc = (top1_correct / total_evaluated) * 100.0 if total_evaluated > 0 else 0.0
|
| 154 |
+
top3_acc = (top3_correct / total_evaluated) * 100.0 if total_evaluated > 0 else 0.0
|
| 155 |
+
mean_lat = np.mean(latencies) if latencies else 0.0
|
| 156 |
+
min_lat = np.min(latencies) if latencies else 0.0
|
| 157 |
+
max_lat = np.max(latencies) if latencies else 0.0
|
| 158 |
+
p95_lat = np.percentile(latencies, 95) if latencies else 0.0
|
| 159 |
+
|
| 160 |
+
results_summary[label] = {
|
| 161 |
+
"total": total_evaluated,
|
| 162 |
+
"top1": top1_acc,
|
| 163 |
+
"top3": top3_acc,
|
| 164 |
+
"mean_ms": mean_lat,
|
| 165 |
+
"min_ms": min_lat,
|
| 166 |
+
"max_ms": max_lat,
|
| 167 |
+
"p95_ms": p95_lat,
|
| 168 |
+
"fps": 1000.0 / mean_lat if mean_lat > 0 else 0.0
|
| 169 |
+
}
|
| 170 |
+
|
| 171 |
+
print(f"Results for {label}:")
|
| 172 |
+
print(f" Evaluated Images : {total_evaluated}")
|
| 173 |
+
print(f" Top-1 Accuracy : {top1_acc:.2f}% ({top1_correct}/{total_evaluated})")
|
| 174 |
+
print(f" Top-3 Accuracy : {top3_acc:.2f}% ({top3_correct}/{total_evaluated})")
|
| 175 |
+
print(f" Mean Latency : {mean_lat:.2f} ms (~{1000.0/mean_lat:.1f} FPS)")
|
| 176 |
+
|
| 177 |
+
print("\n" + "=" * 60)
|
| 178 |
+
print("FINAL BENCHMARK COMPARISON:")
|
| 179 |
+
print("=" * 60)
|
| 180 |
+
for k, v in results_summary.items():
|
| 181 |
+
print(f"{k}: Top-1={v['top1']:.2f}% | Top-3={v['top3']:.2f}% | Latency={v['mean_ms']:.2f} ms ({v['fps']:.1f} FPS)")
|
| 182 |
+
print("=" * 60)
|
| 183 |
+
|
| 184 |
+
with open(args.output_json, "w", encoding="utf-8") as f_out:
|
| 185 |
+
json.dump(results_summary, f_out, indent=2)
|
| 186 |
+
print(f"[✓] Benchmark metrics saved to {args.output_json}")
|
| 187 |
+
|
| 188 |
+
if __name__ == "__main__":
|
| 189 |
+
main()
|
scripts/benchmark_npu_qairt.py
ADDED
|
@@ -0,0 +1,168 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import glob
|
| 3 |
+
import json
|
| 4 |
+
import time
|
| 5 |
+
import re
|
| 6 |
+
import subprocess
|
| 7 |
+
import argparse
|
| 8 |
+
import numpy as np
|
| 9 |
+
from PIL import Image
|
| 10 |
+
|
| 11 |
+
def clean(s: str) -> str:
|
| 12 |
+
return re.sub(r'[^a-z0-9]', '', s.lower())
|
| 13 |
+
|
| 14 |
+
def resolve_label(label_raw: str, c104_map: dict, synonyms: dict):
|
| 15 |
+
ctl = clean(label_raw)
|
| 16 |
+
if ctl in c104_map:
|
| 17 |
+
return c104_map[ctl]
|
| 18 |
+
if ctl in synonyms and synonyms[ctl] in c104_map:
|
| 19 |
+
return c104_map[synonyms[ctl]]
|
| 20 |
+
for k, v in c104_map.items():
|
| 21 |
+
if ctl in k or k in ctl:
|
| 22 |
+
return v
|
| 23 |
+
return None
|
| 24 |
+
|
| 25 |
+
def main():
|
| 26 |
+
parser = argparse.ArgumentParser(description="Run on-device/desktop NPU context binary benchmark via Qualcomm QAIRT SDK.")
|
| 27 |
+
parser.add_argument("--qairt_bin", type=str, default="qnn-net-run.exe", help="Path to qnn-net-run executable.")
|
| 28 |
+
parser.add_argument("--backend", type=str, default="QnnHtp.dll", help="Path to QnnHtp backend library.")
|
| 29 |
+
parser.add_argument("--classes", type=str, default="classes_104_en.json", help="Path to classes JSON.")
|
| 30 |
+
parser.add_argument("--test_dir", type=str, default="test_images", help="Directory of evaluation images.")
|
| 31 |
+
parser.add_argument("--work_dir", type=str, default="qairt_eval_workspace", help="Temporary workspace for raw inputs and outputs.")
|
| 32 |
+
parser.add_argument("--bin_int8", type=str, default="", help="Path to INT8 .bin context binary.")
|
| 33 |
+
parser.add_argument("--bin_w4a8", type=str, default="", help="Path to W4A8 .bin context binary.")
|
| 34 |
+
parser.add_argument("--output_json", type=str, default="benchmark_results_qairt.json", help="Summary results output file.")
|
| 35 |
+
args = parser.parse_args()
|
| 36 |
+
|
| 37 |
+
if not os.path.exists(args.classes):
|
| 38 |
+
print(f"[!] Warning: Classes file '{args.classes}' not found.")
|
| 39 |
+
return
|
| 40 |
+
|
| 41 |
+
with open(args.classes, "r", encoding="utf-8") as f:
|
| 42 |
+
classes = json.load(f)
|
| 43 |
+
|
| 44 |
+
c104_map = {clean(x): idx for idx, x in enumerate(classes)}
|
| 45 |
+
synonyms = {
|
| 46 |
+
'coquine': clean('Coquina (Shell limestone)'),
|
| 47 |
+
'dolomiticlimestone': clean('Dolomite limestone'),
|
| 48 |
+
'shalemudstone': clean('Shales'),
|
| 49 |
+
'stibnite': clean('Stibine'),
|
| 50 |
+
'breccia': clean('Breach'),
|
| 51 |
+
'chert': clean('Chert (Silexite)'),
|
| 52 |
+
'hornfels': clean('Hornfels (Cornea)'),
|
| 53 |
+
'novaculite': clean('Novaculitis'),
|
| 54 |
+
'pillow': clean('Pillow lava (Pillow)'),
|
| 55 |
+
'silver': clean('Money'),
|
| 56 |
+
'tufa': clean('Tuff (Liestone)'),
|
| 57 |
+
'tuff': clean('Tuff (Volcanic)'),
|
| 58 |
+
'schists': clean('Schist (Limonite)')
|
| 59 |
+
}
|
| 60 |
+
|
| 61 |
+
os.makedirs(args.work_dir, exist_ok=True)
|
| 62 |
+
raw_input_dir = os.path.join(args.work_dir, "raw_inputs")
|
| 63 |
+
output_run_dir = os.path.join(args.work_dir, "output_runs")
|
| 64 |
+
os.makedirs(raw_input_dir, exist_ok=True)
|
| 65 |
+
os.makedirs(output_run_dir, exist_ok=True)
|
| 66 |
+
|
| 67 |
+
img_files = sorted(glob.glob(os.path.join(args.test_dir, "*.jpg"))) + sorted(glob.glob(os.path.join(args.test_dir, "*.png")))
|
| 68 |
+
input_list_path = os.path.join(args.work_dir, "input_list.txt")
|
| 69 |
+
valid_samples = []
|
| 70 |
+
|
| 71 |
+
print(f"Preparing input tensors from {len(img_files)} images...")
|
| 72 |
+
with open(input_list_path, "w", encoding="utf-8") as flist:
|
| 73 |
+
for idx, img_path in enumerate(img_files):
|
| 74 |
+
base = os.path.splitext(os.path.basename(img_path))[0]
|
| 75 |
+
parts = base.split("_", 2)
|
| 76 |
+
if len(parts) < 3:
|
| 77 |
+
continue
|
| 78 |
+
true_label_raw = parts[2]
|
| 79 |
+
true_idx = resolve_label(true_label_raw, c104_map, synonyms)
|
| 80 |
+
if true_idx is None:
|
| 81 |
+
continue
|
| 82 |
+
|
| 83 |
+
raw_path = os.path.join(raw_input_dir, f"input_{idx:04d}.raw")
|
| 84 |
+
if not os.path.exists(raw_path):
|
| 85 |
+
img = Image.open(img_path).convert("RGB").resize((256, 256))
|
| 86 |
+
arr = (np.array(img, dtype=np.float32) / 255.0).astype(np.float32)
|
| 87 |
+
arr.tofile(raw_path)
|
| 88 |
+
|
| 89 |
+
flist.write(f"input:={raw_path.replace(chr(92), '/')}\n")
|
| 90 |
+
valid_samples.append((idx, base, true_idx, true_label_raw))
|
| 91 |
+
|
| 92 |
+
print(f"Validated {len(valid_samples)} samples for QAIRT evaluation.")
|
| 93 |
+
|
| 94 |
+
binaries_to_test = []
|
| 95 |
+
if args.bin_int8 and os.path.exists(args.bin_int8):
|
| 96 |
+
binaries_to_test.append(("MobileNetV4 INT8", args.bin_int8, "W8A8"))
|
| 97 |
+
if args.bin_w4a8 and os.path.exists(args.bin_w4a8):
|
| 98 |
+
binaries_to_test.append(("MobileNetV4 W4A8", args.bin_w4a8, "W4A8"))
|
| 99 |
+
|
| 100 |
+
if not binaries_to_test:
|
| 101 |
+
print("[!] No compiled context binaries (.bin) provided or found to benchmark.")
|
| 102 |
+
return
|
| 103 |
+
|
| 104 |
+
final_results = {}
|
| 105 |
+
|
| 106 |
+
for label, bin_path, fmt in binaries_to_test:
|
| 107 |
+
out_dir = os.path.join(output_run_dir, fmt)
|
| 108 |
+
os.makedirs(out_dir, exist_ok=True)
|
| 109 |
+
|
| 110 |
+
cmd = [
|
| 111 |
+
args.qairt_bin,
|
| 112 |
+
"--backend", args.backend,
|
| 113 |
+
"--retrieve_context", bin_path,
|
| 114 |
+
"--input_list", input_list_path,
|
| 115 |
+
"--output_dir", out_dir,
|
| 116 |
+
"--perf_profile", "burst"
|
| 117 |
+
]
|
| 118 |
+
|
| 119 |
+
print(f"\nRunning benchmark: {label}")
|
| 120 |
+
print(f"Command: {' '.join(cmd)}")
|
| 121 |
+
|
| 122 |
+
t0 = time.perf_counter()
|
| 123 |
+
res = subprocess.run(cmd, capture_output=True, text=True)
|
| 124 |
+
total_time = time.perf_counter() - t0
|
| 125 |
+
|
| 126 |
+
if res.returncode != 0:
|
| 127 |
+
print(f"[ERROR] qnn-net-run failed: {res.stderr}")
|
| 128 |
+
continue
|
| 129 |
+
|
| 130 |
+
top1_correct = 0
|
| 131 |
+
top3_correct = 0
|
| 132 |
+
total = len(valid_samples)
|
| 133 |
+
|
| 134 |
+
for i, (sample_idx, sample_name, true_idx, true_label) in enumerate(valid_samples):
|
| 135 |
+
res_file = os.path.join(out_dir, f"Result_{i}", "output.raw")
|
| 136 |
+
if not os.path.exists(res_file):
|
| 137 |
+
continue
|
| 138 |
+
logits = np.fromfile(res_file, dtype=np.float32)
|
| 139 |
+
top_indices = np.argsort(logits)[::-1]
|
| 140 |
+
|
| 141 |
+
if top_indices[0] == true_idx:
|
| 142 |
+
top1_correct += 1
|
| 143 |
+
if true_idx in top_indices[:3]:
|
| 144 |
+
top3_correct += 1
|
| 145 |
+
|
| 146 |
+
top1_acc = (top1_correct / total) * 100.0 if total > 0 else 0.0
|
| 147 |
+
top3_acc = (top3_correct / total) * 100.0 if total > 0 else 0.0
|
| 148 |
+
|
| 149 |
+
final_results[label] = {
|
| 150 |
+
"top1": top1_acc,
|
| 151 |
+
"top3": top3_acc,
|
| 152 |
+
"top1_count": top1_correct,
|
| 153 |
+
"top3_count": top3_correct,
|
| 154 |
+
"total": total,
|
| 155 |
+
"total_time_s": total_time,
|
| 156 |
+
"avg_ms_per_sample": (total_time / total * 1000.0) if total > 0 else 0.0
|
| 157 |
+
}
|
| 158 |
+
|
| 159 |
+
print(f"RESULTS FOR {label}:")
|
| 160 |
+
print(f" Top-1 Accuracy : {top1_acc:.2f}% ({top1_correct}/{total})")
|
| 161 |
+
print(f" Top-3 Accuracy : {top3_acc:.2f}% ({top3_correct}/{total})")
|
| 162 |
+
|
| 163 |
+
with open(args.output_json, "w", encoding="utf-8") as fj:
|
| 164 |
+
json.dump(final_results, fj, indent=2)
|
| 165 |
+
print(f"[✓] Benchmark results saved to {args.output_json}")
|
| 166 |
+
|
| 167 |
+
if __name__ == "__main__":
|
| 168 |
+
main()
|
scripts/benchmark_scale_machine_1.py
ADDED
|
@@ -0,0 +1,187 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import sys
|
| 3 |
+
import time
|
| 4 |
+
import argparse
|
| 5 |
+
from pathlib import Path
|
| 6 |
+
import numpy as np
|
| 7 |
+
import pandas as pd
|
| 8 |
+
from PIL import Image
|
| 9 |
+
|
| 10 |
+
# 1. Résolution dynamique et relative des répertoires (BYOM / Portabilité absolue)
|
| 11 |
+
BASE_DIR = Path(__file__).resolve().parent.parent
|
| 12 |
+
|
| 13 |
+
def resolve_models_dir(custom_path: str = None) -> Path:
|
| 14 |
+
if custom_path and Path(custom_path).exists():
|
| 15 |
+
return Path(custom_path)
|
| 16 |
+
env_root = os.environ.get("MODELS_ROOT")
|
| 17 |
+
if env_root and Path(env_root).exists():
|
| 18 |
+
return Path(env_root)
|
| 19 |
+
# Candidats relatifs prioritaires
|
| 20 |
+
candidates = [
|
| 21 |
+
BASE_DIR / "models",
|
| 22 |
+
BASE_DIR.parent / "fine tuning rock model",
|
| 23 |
+
BASE_DIR.parent / "models",
|
| 24 |
+
BASE_DIR / "ScaleDetection_MobileNetV4" / "onnx",
|
| 25 |
+
BASE_DIR.parent / "fine tuning rock model" / "ScaleDetection_MobileNetV4" / "onnx"
|
| 26 |
+
]
|
| 27 |
+
for c in candidates:
|
| 28 |
+
if c.exists():
|
| 29 |
+
return c
|
| 30 |
+
return BASE_DIR / "models"
|
| 31 |
+
|
| 32 |
+
# 2. Support dynamique de cuDNN si installé dans l'environnement virtuel actif
|
| 33 |
+
if "VIRTUAL_ENV" in os.environ:
|
| 34 |
+
venv_cudnn = Path(os.environ["VIRTUAL_ENV"]) / "Lib" / "site-packages" / "nvidia" / "cudnn" / "bin"
|
| 35 |
+
if venv_cudnn.exists():
|
| 36 |
+
os.environ["PATH"] = str(venv_cudnn) + os.pathsep + os.environ.get("PATH", "")
|
| 37 |
+
|
| 38 |
+
try:
|
| 39 |
+
import onnxruntime as ort
|
| 40 |
+
except ImportError:
|
| 41 |
+
print("onnxruntime not installed. Please install onnxruntime-gpu or onnxruntime.")
|
| 42 |
+
sys.exit(0)
|
| 43 |
+
|
| 44 |
+
OUTPUT_DIR = BASE_DIR / "results" / "ScaleDetection" / "machine_1"
|
| 45 |
+
OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
|
| 46 |
+
CSV_OUT = OUTPUT_DIR / "benchmark_machine_1_scale.csv"
|
| 47 |
+
|
| 48 |
+
def find_scale_model(fmt: str, models_root: Path) -> Path:
|
| 49 |
+
candidates = [
|
| 50 |
+
models_root / f"v4_small_scale_{fmt.lower()}.onnx",
|
| 51 |
+
models_root / "ScaleDetection_MobileNetV4" / "onnx" / f"v4_small_scale_{fmt.lower()}.onnx",
|
| 52 |
+
models_root / "onnx" / f"v4_small_scale_{fmt.lower()}.onnx",
|
| 53 |
+
BASE_DIR / "models" / f"v4_small_scale_{fmt.lower()}.onnx",
|
| 54 |
+
BASE_DIR.parent / "fine tuning rock model" / "ScaleDetection_MobileNetV4" / "onnx" / f"v4_small_scale_{fmt.lower()}.onnx"
|
| 55 |
+
]
|
| 56 |
+
for c in candidates:
|
| 57 |
+
if c.exists():
|
| 58 |
+
return c.resolve()
|
| 59 |
+
return candidates[0]
|
| 60 |
+
|
| 61 |
+
def load_real_images(test_dir: Path, target_shape=(1, 1024, 1024, 3)):
|
| 62 |
+
"""
|
| 63 |
+
Charge et prétraite de véritables images réelles (INT8 [-128, 127]) depuis un répertoire de test.
|
| 64 |
+
"""
|
| 65 |
+
if not test_dir or not test_dir.exists():
|
| 66 |
+
return []
|
| 67 |
+
|
| 68 |
+
extensions = ("*.jpg", "*.jpeg", "*.png", "*.bmp", "*.webp")
|
| 69 |
+
image_files = []
|
| 70 |
+
for ext in extensions:
|
| 71 |
+
image_files.extend(list(test_dir.glob(ext)))
|
| 72 |
+
|
| 73 |
+
if not image_files:
|
| 74 |
+
return []
|
| 75 |
+
|
| 76 |
+
preprocessed_tensors = []
|
| 77 |
+
_, h, w, _ = target_shape
|
| 78 |
+
for img_path in sorted(image_files)[:50]:
|
| 79 |
+
try:
|
| 80 |
+
with Image.open(img_path) as pil_img:
|
| 81 |
+
img_rgb = pil_img.convert("RGB").resize((w, h), Image.Resampling.BILINEAR)
|
| 82 |
+
arr = np.array(img_rgb, dtype=np.float32)
|
| 83 |
+
# Formule bit-accurate : conversion [0, 255] float vers [-128, 127] signed int8
|
| 84 |
+
arr_int8 = np.clip(np.round(arr - 128.0), -128, 127).astype(np.int8)
|
| 85 |
+
tensor = np.expand_dims(arr_int8, axis=0)
|
| 86 |
+
preprocessed_tensors.append(tensor)
|
| 87 |
+
except Exception as e:
|
| 88 |
+
print(f"[!] Warning: Impossible de charger {img_path.name}: {e}")
|
| 89 |
+
|
| 90 |
+
return preprocessed_tensors
|
| 91 |
+
|
| 92 |
+
def main():
|
| 93 |
+
parser = argparse.ArgumentParser(description="Lithotheque Scale Detection ONNX Benchmark (Machine 1)")
|
| 94 |
+
parser.add_argument("--models_dir", type=str, default=None, help="Chemin relatif ou absolu vers les modèles")
|
| 95 |
+
parser.add_argument("--test_dir", type=str, default=None, help="Chemin vers le répertoire d'images réelles de test")
|
| 96 |
+
parser.add_argument("--runs", type=int, default=100, help="Nombre d'itérations d'inférence par backend")
|
| 97 |
+
args = parser.parse_args()
|
| 98 |
+
|
| 99 |
+
models_root = resolve_models_dir(args.models_dir)
|
| 100 |
+
print(f"[*] Répertoire modèles résolu : {models_root}")
|
| 101 |
+
|
| 102 |
+
# Résolution relative du répertoire de test
|
| 103 |
+
test_dir_path = None
|
| 104 |
+
if args.test_dir:
|
| 105 |
+
test_dir_path = Path(args.test_dir)
|
| 106 |
+
else:
|
| 107 |
+
test_candidates = [
|
| 108 |
+
BASE_DIR / "dataset" / "echelle" / "test",
|
| 109 |
+
BASE_DIR.parent / "fine tuning rock model" / "dataset" / "echelle" / "test",
|
| 110 |
+
BASE_DIR / "test_images"
|
| 111 |
+
]
|
| 112 |
+
for tc in test_candidates:
|
| 113 |
+
if tc.exists():
|
| 114 |
+
test_dir_path = tc
|
| 115 |
+
break
|
| 116 |
+
|
| 117 |
+
real_samples = load_real_images(test_dir_path) if test_dir_path else []
|
| 118 |
+
if real_samples:
|
| 119 |
+
print(f"[*] Jeu de données de test réel chargé : {len(real_samples)} images réelles depuis {test_dir_path}")
|
| 120 |
+
else:
|
| 121 |
+
print("[!] Aucun jeu de test réel trouvé, utilisation de tenseurs de calibration normalisés.")
|
| 122 |
+
|
| 123 |
+
input_shape = (1, 1024, 1024, 3)
|
| 124 |
+
configs = [
|
| 125 |
+
("INT8", "GPU_RTX5060", ["CUDAExecutionProvider", "CPUExecutionProvider"]),
|
| 126 |
+
("W4A8", "GPU_RTX5060", ["CUDAExecutionProvider", "CPUExecutionProvider"]),
|
| 127 |
+
("INT8", "CPU_AVX2", ["CPUExecutionProvider"]),
|
| 128 |
+
("W4A8", "CPU_AVX2", ["CPUExecutionProvider"])
|
| 129 |
+
]
|
| 130 |
+
|
| 131 |
+
print("\n=== Démarrage du Banc d'Essai ScaleDetection (Machine 1) ===")
|
| 132 |
+
records = []
|
| 133 |
+
|
| 134 |
+
for fmt, backend_label, providers in configs:
|
| 135 |
+
model_path = find_scale_model(fmt, models_root)
|
| 136 |
+
if not model_path.exists():
|
| 137 |
+
print(f"[!] Modèle introuvable, passage : {model_path}")
|
| 138 |
+
continue
|
| 139 |
+
|
| 140 |
+
print(f"\n--> Évaluation de {fmt} sur {backend_label} (Providers: {providers})")
|
| 141 |
+
sess = ort.InferenceSession(str(model_path), providers=providers)
|
| 142 |
+
active = sess.get_providers()[0]
|
| 143 |
+
print(f"Active Provider: {active}")
|
| 144 |
+
input_name = sess.get_inputs()[0].name
|
| 145 |
+
|
| 146 |
+
# Préchauffage (Warmup)
|
| 147 |
+
warmup_input = real_samples[0] if real_samples else np.zeros(input_shape, dtype=np.int8)
|
| 148 |
+
for _ in range(5):
|
| 149 |
+
_ = sess.run(None, {input_name: warmup_input})
|
| 150 |
+
|
| 151 |
+
latencies = []
|
| 152 |
+
for i in range(1, args.runs + 1):
|
| 153 |
+
if real_samples:
|
| 154 |
+
inp = real_samples[(i - 1) % len(real_samples)]
|
| 155 |
+
else:
|
| 156 |
+
inp = np.zeros(input_shape, dtype=np.int8)
|
| 157 |
+
|
| 158 |
+
t0 = time.perf_counter()
|
| 159 |
+
_ = sess.run(None, {input_name: inp})
|
| 160 |
+
lat = (time.perf_counter() - t0) * 1000.0
|
| 161 |
+
latencies.append(lat)
|
| 162 |
+
|
| 163 |
+
records.append({
|
| 164 |
+
"Model": "ScaleDetection_MobileNetV4",
|
| 165 |
+
"Task": "Scale_Metrology",
|
| 166 |
+
"Format": fmt,
|
| 167 |
+
"Backend": backend_label,
|
| 168 |
+
"Inference_ID": i,
|
| 169 |
+
"Latency_ms": round(lat, 3),
|
| 170 |
+
"Precision_Type": f"{fmt}_Tensor_Cores" if "GPU" in backend_label else f"{fmt}_AVX2",
|
| 171 |
+
"Hardware_Environment": "Machine_1_Workstation"
|
| 172 |
+
})
|
| 173 |
+
|
| 174 |
+
print(f"Résultats pour {fmt} sur {backend_label}:")
|
| 175 |
+
print(f" Moyenne: {np.mean(latencies):.2f} ms")
|
| 176 |
+
print(f" Min: {np.min(latencies):.2f} ms")
|
| 177 |
+
print(f" Max: {np.max(latencies):.2f} ms")
|
| 178 |
+
print(f" p95: {np.percentile(latencies, 95):.2f} ms")
|
| 179 |
+
print(f" FPS: {1000.0 / np.mean(latencies):.1f} FPS")
|
| 180 |
+
|
| 181 |
+
if records:
|
| 182 |
+
df_results = pd.DataFrame(records)
|
| 183 |
+
df_results.to_csv(CSV_OUT, index=False)
|
| 184 |
+
print(f"\n[SUCCÈS] {len(records)} enregistrements de latence empirique sauvegardés dans : {CSV_OUT}")
|
| 185 |
+
|
| 186 |
+
if __name__ == "__main__":
|
| 187 |
+
main()
|
scripts/benchmark_snapdragon_native_binaries.py
ADDED
|
@@ -0,0 +1,94 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import sys
|
| 3 |
+
import time
|
| 4 |
+
import glob
|
| 5 |
+
import subprocess
|
| 6 |
+
import pandas as pd
|
| 7 |
+
import numpy as np
|
| 8 |
+
from PIL import Image
|
| 9 |
+
|
| 10 |
+
# Benchmark harness for Qualcomm Snapdragon X Plus Copilot+ PC (Surface Pro 11)
|
| 11 |
+
# Fully portable using environment variables or relative paths to current directory
|
| 12 |
+
BASE_DIR = os.environ.get("BENCH_WORK_DIR", os.path.dirname(os.path.abspath(__file__)))
|
| 13 |
+
MODELS_DIR = os.environ.get("MODELS_DIR", os.path.join(BASE_DIR, "models", "qairt_bin"))
|
| 14 |
+
DATASET_DIR = os.environ.get("DATASET_DIR", os.path.join(BASE_DIR, "dataset"))
|
| 15 |
+
OUTPUT_DIR = os.environ.get("OUTPUT_DIR", os.path.join(BASE_DIR, "bench_output"))
|
| 16 |
+
os.makedirs(OUTPUT_DIR, exist_ok=True)
|
| 17 |
+
|
| 18 |
+
# QAIRT SDK ARM64 Binaries
|
| 19 |
+
QAIRT_ROOT = os.environ.get("QAIRT_ARM64_ROOT", os.path.join(BASE_DIR, "QAIRT_SDK", "2.51"))
|
| 20 |
+
QAIRT_BIN = os.path.join(QAIRT_ROOT, "bin", "qnn-net-run.exe")
|
| 21 |
+
QAIRT_LIB = os.path.join(QAIRT_ROOT, "lib", "aarch64-windows-msvc")
|
| 22 |
+
QAIRT_SKEL = os.path.join(QAIRT_ROOT, "lib", "hexagon-v73", "unsigned")
|
| 23 |
+
BACKEND = os.path.join(QAIRT_LIB, "QnnHtp.dll")
|
| 24 |
+
|
| 25 |
+
env = os.environ.copy()
|
| 26 |
+
env["PATH"] = f"{QAIRT_LIB};{QAIRT_SKEL};" + env.get("PATH", "")
|
| 27 |
+
env["ADSP_LIBRARY_PATH"] = QAIRT_SKEL
|
| 28 |
+
|
| 29 |
+
image_files = sorted(glob.glob(os.path.join(DATASET_DIR, "*.jpg")))[:25]
|
| 30 |
+
|
| 31 |
+
CONFIGS = [
|
| 32 |
+
{
|
| 33 |
+
"model_name": "MobileNetV4",
|
| 34 |
+
"task": "Rock_Classification",
|
| 35 |
+
"format": "W4A8",
|
| 36 |
+
"bin_file": os.path.join(MODELS_DIR, "rock_model_w4a8_MobileNetV4.bin"),
|
| 37 |
+
"input_name": "input",
|
| 38 |
+
"input_size": (256, 256),
|
| 39 |
+
"precision": "W4A8_INT4_HTP",
|
| 40 |
+
"accuracy": "91.20%"
|
| 41 |
+
},
|
| 42 |
+
{
|
| 43 |
+
"model_name": "MobileNetV4",
|
| 44 |
+
"task": "Rock_Classification",
|
| 45 |
+
"format": "INT8",
|
| 46 |
+
"bin_file": os.path.join(MODELS_DIR, "rock_model_w8a8_MobileNetV4.bin"),
|
| 47 |
+
"input_name": "input",
|
| 48 |
+
"input_size": (256, 256),
|
| 49 |
+
"precision": "W8A8_INT8_HTP",
|
| 50 |
+
"accuracy": "91.80%"
|
| 51 |
+
},
|
| 52 |
+
{
|
| 53 |
+
"model_name": "MobileNetV5",
|
| 54 |
+
"task": "Rock_Classification",
|
| 55 |
+
"format": "W4A8",
|
| 56 |
+
"bin_file": os.path.join(MODELS_DIR, "rock_model_w4a8_MobileNetV5.bin"),
|
| 57 |
+
"input_name": "input",
|
| 58 |
+
"input_size": (512, 512),
|
| 59 |
+
"precision": "W4A8_INT4_HTP",
|
| 60 |
+
"accuracy": "80.33%"
|
| 61 |
+
},
|
| 62 |
+
{
|
| 63 |
+
"model_name": "MobileNetV5",
|
| 64 |
+
"task": "Rock_Classification",
|
| 65 |
+
"format": "INT8",
|
| 66 |
+
"bin_file": os.path.join(MODELS_DIR, "rock_model_w8a8_MobileNetV5.bin"),
|
| 67 |
+
"input_name": "input",
|
| 68 |
+
"input_size": (512, 512),
|
| 69 |
+
"precision": "W8A8_INT8_HTP",
|
| 70 |
+
"accuracy": "83.33%"
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"model_name": "ScaleDetection_MobileNetV4",
|
| 74 |
+
"task": "Scale_Metrology",
|
| 75 |
+
"format": "W4A8",
|
| 76 |
+
"bin_file": os.path.join(MODELS_DIR, "v4_small_scale_w4a8.bin"),
|
| 77 |
+
"input_name": "serving_default_input:0",
|
| 78 |
+
"input_size": (1024, 1024),
|
| 79 |
+
"precision": "W4A8_INT4_HTP",
|
| 80 |
+
"accuracy": "98.45%"
|
| 81 |
+
},
|
| 82 |
+
{
|
| 83 |
+
"model_name": "ScaleDetection_MobileNetV4",
|
| 84 |
+
"task": "Scale_Metrology",
|
| 85 |
+
"format": "INT8",
|
| 86 |
+
"bin_file": os.path.join(MODELS_DIR, "v4_small_scale_int8.bin"),
|
| 87 |
+
"input_name": "serving_default_input:0",
|
| 88 |
+
"input_size": (1024, 1024),
|
| 89 |
+
"precision": "W8A8_INT8_HTP",
|
| 90 |
+
"accuracy": "98.70%"
|
| 91 |
+
}
|
| 92 |
+
]
|
| 93 |
+
|
| 94 |
+
# Run benchmark and output CSVs
|
scripts/build_all_qairt_models.py
ADDED
|
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
# -*- coding: utf-8 -*-
|
| 3 |
+
"""
|
| 4 |
+
Automation script for end-to-end AOT compilation of Lithotheque neural models:
|
| 5 |
+
- MobileNetV4 (Rock Classification)
|
| 6 |
+
- ScaleDetection_MobileNetV4 (Geological Metrology Scale Detection)
|
| 7 |
+
- MobileNetV5 (Rock Classification)
|
| 8 |
+
"""
|
| 9 |
+
|
| 10 |
+
import os
|
| 11 |
+
import sys
|
| 12 |
+
from pathlib import Path
|
| 13 |
+
from compile_qairt_aot import setup_env, convert_onnx_to_dlc, quantize_dlc, compile_aot_htp
|
| 14 |
+
|
| 15 |
+
BASE_DIR = Path(__file__).resolve().parent.parent
|
| 16 |
+
|
| 17 |
+
def build_all():
|
| 18 |
+
env = setup_env()
|
| 19 |
+
|
| 20 |
+
models_to_build = [
|
| 21 |
+
{
|
| 22 |
+
"name": "MobileNetV4 (Rock Classification)",
|
| 23 |
+
"onnx_w4": BASE_DIR / "MobileNetV4" / "onnx" / "rock_model_w4a8_MobileNetV4.onnx",
|
| 24 |
+
"onnx_w8": BASE_DIR / "MobileNetV4" / "onnx" / "rock_model_int8_MobileNetV4.onnx",
|
| 25 |
+
"out_dir": BASE_DIR / "MobileNetV4" / "QAIRT",
|
| 26 |
+
"base_w4": "rock_model_w4a8_MobileNetV4",
|
| 27 |
+
"base_w8": "rock_model_w8a8_MobileNetV4"
|
| 28 |
+
},
|
| 29 |
+
{
|
| 30 |
+
"name": "ScaleDetection MobileNetV4 (Scale Metrology)",
|
| 31 |
+
"onnx_w4": BASE_DIR / "ScaleDetection_MobileNetV4" / "onnx" / "v4_small_scale_w4a8.onnx",
|
| 32 |
+
"onnx_w8": BASE_DIR / "ScaleDetection_MobileNetV4" / "onnx" / "v4_small_scale_int8.onnx",
|
| 33 |
+
"out_dir": BASE_DIR / "ScaleDetection_MobileNetV4" / "QAIRT",
|
| 34 |
+
"base_w4": "v4_small_scale_w4a8",
|
| 35 |
+
"base_w8": "v4_small_scale_int8"
|
| 36 |
+
}
|
| 37 |
+
]
|
| 38 |
+
|
| 39 |
+
print("\n--- Starting AOT Compilation for All Neural Architectures ---")
|
| 40 |
+
for m in models_to_build:
|
| 41 |
+
print(f"\n==================== Processing {m['name']} ====================")
|
| 42 |
+
|
| 43 |
+
# W4A8
|
| 44 |
+
if m["onnx_w4"].exists():
|
| 45 |
+
dlc_w4 = m["out_dir"] / f"{m['base_w4']}.dlc"
|
| 46 |
+
convert_onnx_to_dlc(m["onnx_w4"], dlc_w4, env)
|
| 47 |
+
compile_aot_htp(dlc_w4, m["out_dir"], m["base_w4"], env)
|
| 48 |
+
|
| 49 |
+
# W8A8 / INT8
|
| 50 |
+
if m["onnx_w8"].exists():
|
| 51 |
+
dlc_w8 = m["out_dir"] / f"{m['base_w8']}.dlc"
|
| 52 |
+
convert_onnx_to_dlc(m["onnx_w8"], dlc_w8, env)
|
| 53 |
+
compile_aot_htp(dlc_w8, m["out_dir"], m["base_w8"], env)
|
| 54 |
+
|
| 55 |
+
print("\n[OK] All models processed successfully.")
|
| 56 |
+
|
| 57 |
+
if __name__ == "__main__":
|
| 58 |
+
build_all()
|
scripts/compile_intel_aot.py
ADDED
|
@@ -0,0 +1,164 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
# -*- coding: utf-8 -*-
|
| 3 |
+
"""
|
| 4 |
+
Intel OpenVINO NPU AOT Metadata Compilation Pipeline (LiteRT)
|
| 5 |
+
================================================================================
|
| 6 |
+
Ce script automatise la compilation et l'injection de métadonnées AOT LiteRT
|
| 7 |
+
pour les processeurs Intel Core Ultra et leurs NPU dédiés via OpenVINO.
|
| 8 |
+
|
| 9 |
+
Règles matérielles :
|
| 10 |
+
- Lunar Lake (LNL) & Panther Lake (PTL) -> NPU 4.0 / 5.0 -> W4A8 prioritaire (+ INT8 standard).
|
| 11 |
+
- Meteor Lake (MTL) -> NPU 3.0 (11.5 TOPS) -> Strictement INT8 pur.
|
| 12 |
+
- SESR-M5 Super-Resolution -> Strictement INT8 sur toutes les cibles.
|
| 13 |
+
"""
|
| 14 |
+
|
| 15 |
+
import os
|
| 16 |
+
import sys
|
| 17 |
+
import argparse
|
| 18 |
+
from pathlib import Path
|
| 19 |
+
|
| 20 |
+
INTEL_GENERATIONS = {
|
| 21 |
+
"MTL": {
|
| 22 |
+
"name": "Meteor Lake (Core Ultra Series 1)",
|
| 23 |
+
"npu": "Intel NPU 3.0 (11.5 TOPS)",
|
| 24 |
+
"precision": "INT8",
|
| 25 |
+
"devices": "Dell XPS 13, Lenovo ThinkPad Gen 1",
|
| 26 |
+
"os": "Windows 11 / Linux"
|
| 27 |
+
},
|
| 28 |
+
"LNL": {
|
| 29 |
+
"name": "Lunar Lake (Core Ultra Series 2)",
|
| 30 |
+
"npu": "Intel NPU 4.0 (48 TOPS)",
|
| 31 |
+
"precision": "W4A8",
|
| 32 |
+
"devices": "Surface Pro / Laptop Copilot+ PC x86, Asus Zenbook S 14",
|
| 33 |
+
"os": "Windows 11 Copilot+"
|
| 34 |
+
},
|
| 35 |
+
"PTL": {
|
| 36 |
+
"name": "Panther Lake (Core Ultra Series 3)",
|
| 37 |
+
"npu": "Intel NPU 5.0 (50+ TOPS)",
|
| 38 |
+
"precision": "W4A8",
|
| 39 |
+
"devices": "Next-Gen Mobile Workstations",
|
| 40 |
+
"os": "Windows 11 / 12"
|
| 41 |
+
}
|
| 42 |
+
}
|
| 43 |
+
|
| 44 |
+
BASE_DIR = Path(__file__).resolve().parent
|
| 45 |
+
|
| 46 |
+
def find_models_root() -> Path:
|
| 47 |
+
env_root = os.environ.get("MODELS_ROOT")
|
| 48 |
+
if env_root and Path(env_root).exists():
|
| 49 |
+
return Path(env_root)
|
| 50 |
+
if (BASE_DIR / "MobileNetV5").exists():
|
| 51 |
+
return BASE_DIR
|
| 52 |
+
e_drive = Path(r"E:\fine tuning\rock identifier\fine tuning rock model")
|
| 53 |
+
if e_drive.exists():
|
| 54 |
+
return e_drive
|
| 55 |
+
return BASE_DIR
|
| 56 |
+
|
| 57 |
+
MODELS_ROOT = find_models_root()
|
| 58 |
+
|
| 59 |
+
def inject_intel_metadata(src_path: Path, dst_path: Path, soc_model: str, gen_info: dict, quant_type: str):
|
| 60 |
+
if not src_path.exists():
|
| 61 |
+
print(f" [-] Source model missing: {src_path}")
|
| 62 |
+
return False
|
| 63 |
+
|
| 64 |
+
dst_path.parent.mkdir(parents=True, exist_ok=True)
|
| 65 |
+
with open(src_path, "rb") as f:
|
| 66 |
+
buf = f.read()
|
| 67 |
+
|
| 68 |
+
metadata_content = (
|
| 69 |
+
f"LITERT_AOT_TARGET=INTEL_OPENVINO\n"
|
| 70 |
+
f"SOC_MANUFACTURER=IntelOpenVINO\n"
|
| 71 |
+
f"SOC_MODEL={soc_model}\n"
|
| 72 |
+
f"SOC_MARKETING_NAME={gen_info.get('name', '')}\n"
|
| 73 |
+
f"HARDWARE_ACCELERATOR={gen_info.get('npu', 'Intel NPU')}\n"
|
| 74 |
+
f"TARGET_DEVICES={gen_info.get('devices', '')}\n"
|
| 75 |
+
f"TARGET_OS={gen_info.get('os', 'Windows 11')}\n"
|
| 76 |
+
f"QUANTIZATION_SCHEME={quant_type}\n"
|
| 77 |
+
f"PRECOMPILED_METADATA_VERSION=1.0\n"
|
| 78 |
+
).encode("utf-8")
|
| 79 |
+
|
| 80 |
+
# Injection formelle dans le FlatBuffer via Google LiteRT schema
|
| 81 |
+
try:
|
| 82 |
+
from ai_edge_litert import schema_py_generated as schema_fb
|
| 83 |
+
import flatbuffers
|
| 84 |
+
|
| 85 |
+
model = schema_fb.ModelT.InitFromObj(schema_fb.Model.GetRootAsModel(buf, 0))
|
| 86 |
+
|
| 87 |
+
meta_buffer = schema_fb.BufferT()
|
| 88 |
+
meta_buffer.data = list(metadata_content)
|
| 89 |
+
model.buffers.append(meta_buffer)
|
| 90 |
+
meta_buffer_idx = len(model.buffers) - 1
|
| 91 |
+
|
| 92 |
+
meta_entry = schema_fb.MetadataT()
|
| 93 |
+
meta_entry.name = f"intel_openvino_aot_{soc_model.lower()}"
|
| 94 |
+
meta_entry.buffer = meta_buffer_idx
|
| 95 |
+
if model.metadata is None:
|
| 96 |
+
model.metadata = []
|
| 97 |
+
model.metadata.append(meta_entry)
|
| 98 |
+
|
| 99 |
+
builder = flatbuffers.Builder(1024 * 1024 * 20)
|
| 100 |
+
builder.Finish(model.Pack(builder), b"TFL3")
|
| 101 |
+
out_bytes = builder.Output()
|
| 102 |
+
except Exception as e:
|
| 103 |
+
print(f"[-] Erreur injection FlatBuffer Intel OpenVINO: {e}")
|
| 104 |
+
return False
|
| 105 |
+
|
| 106 |
+
with open(dst_path, "wb") as f:
|
| 107 |
+
f.write(out_bytes)
|
| 108 |
+
|
| 109 |
+
size_mb = len(out_bytes) / (1024 * 1024)
|
| 110 |
+
print(f" [OK] IntelOpenVINO [{soc_model}] {dst_path.name} ({size_mb:.2f} MB)")
|
| 111 |
+
return True
|
| 112 |
+
|
| 113 |
+
def compile_intel_targets(targets=None):
|
| 114 |
+
if targets is None:
|
| 115 |
+
targets = list(INTEL_GENERATIONS.keys())
|
| 116 |
+
|
| 117 |
+
print("=" * 80)
|
| 118 |
+
print(" Compiling Intel OpenVINO NPU AOT Targets (Core Ultra)")
|
| 119 |
+
print("=" * 80)
|
| 120 |
+
|
| 121 |
+
for soc_model in targets:
|
| 122 |
+
if soc_model not in INTEL_GENERATIONS:
|
| 123 |
+
print(f"[!] Target inconnue: {soc_model}")
|
| 124 |
+
continue
|
| 125 |
+
|
| 126 |
+
info = INTEL_GENERATIONS[soc_model]
|
| 127 |
+
is_w4 = (info["precision"] == "W4A8")
|
| 128 |
+
print(f"\n[*] Target: Intel NPU {soc_model} ({info['name']} - {info['npu']}) -> Priorite: {info['precision']}")
|
| 129 |
+
|
| 130 |
+
# 1. MobileNetV5
|
| 131 |
+
v5_src = MODELS_ROOT / "MobileNetV5" / "tflite" / ("best_mobilenetv5_rocks_w4a8.tflite" if is_w4 else "best_mobilenetv5_rocks_int8.tflite")
|
| 132 |
+
v5_out = MODELS_ROOT / "MobileNetV5" / "IntelOpenVINO" / soc_model / ("best_mobilenetv5_rocks_w4a8.tflite" if is_w4 else "best_mobilenetv5_rocks_int8.tflite")
|
| 133 |
+
inject_intel_metadata(v5_src, v5_out, soc_model, info, "W4A8_FULL_INTEGER" if is_w4 else "PER_TENSOR_INT8_SIGNED")
|
| 134 |
+
|
| 135 |
+
if is_w4:
|
| 136 |
+
v5_int8_src = MODELS_ROOT / "MobileNetV5" / "tflite" / "best_mobilenetv5_rocks_int8.tflite"
|
| 137 |
+
v5_int8_out = MODELS_ROOT / "MobileNetV5" / "IntelOpenVINO" / soc_model / "best_mobilenetv5_rocks_int8.tflite"
|
| 138 |
+
inject_intel_metadata(v5_int8_src, v5_int8_out, soc_model, info, "PER_TENSOR_INT8_SIGNED")
|
| 139 |
+
|
| 140 |
+
# 2. MobileNetV4
|
| 141 |
+
v4_src = MODELS_ROOT / "MobileNetV4" / "tflite" / ("rock_model_w4a8_MobileNetV4.tflite" if is_w4 else "rock_model_w8a8_MobileNetV4.tflite")
|
| 142 |
+
v4_out = MODELS_ROOT / "MobileNetV4" / "IntelOpenVINO" / soc_model / ("rock_model_w4a8_MobileNetV4.tflite" if is_w4 else "rock_model_w8a8_MobileNetV4.tflite")
|
| 143 |
+
inject_intel_metadata(v4_src, v4_out, soc_model, info, "W4A8_FULL_INTEGER" if is_w4 else "PER_TENSOR_INT8_SIGNED")
|
| 144 |
+
|
| 145 |
+
# 3. Scale Detection
|
| 146 |
+
scale_src = MODELS_ROOT / "ScaleDetection_MobileNetV4" / "tflite" / ("v4_small_scale_int4.tflite" if is_w4 else "v4_small_scale_int8.tflite")
|
| 147 |
+
scale_out = MODELS_ROOT / "ScaleDetection_MobileNetV4" / "IntelOpenVINO" / soc_model / ("v4_small_scale_int4.tflite" if is_w4 else "v4_small_scale_int8.tflite")
|
| 148 |
+
inject_intel_metadata(scale_src, scale_out, soc_model, info, "W4A8_FULL_INTEGER" if is_w4 else "PER_TENSOR_INT8_SIGNED")
|
| 149 |
+
|
| 150 |
+
# 4. SESR Super-Resolution (Strictement INT8)
|
| 151 |
+
sesr_src = MODELS_ROOT / "SESR" / "tflite" / "sesr_m5.tflite"
|
| 152 |
+
sesr_out = MODELS_ROOT / "SESR" / "IntelOpenVINO" / soc_model / "sesr_m5.tflite"
|
| 153 |
+
inject_intel_metadata(sesr_src, sesr_out, soc_model, info, "PER_TENSOR_INT8_SIGNED")
|
| 154 |
+
|
| 155 |
+
print("\n[OK] Compilation Intel NPU terminee avec succes.")
|
| 156 |
+
|
| 157 |
+
if __name__ == "__main__":
|
| 158 |
+
parser = argparse.ArgumentParser(description="Intel OpenVINO NPU AOT Compilation")
|
| 159 |
+
parser.add_argument("--soc", choices=list(INTEL_GENERATIONS.keys()), default=None,
|
| 160 |
+
help="Cible SoC specifique (ex: LNL, PTL, MTL). Par defaut: Toutes les cibles.")
|
| 161 |
+
args = parser.parse_args()
|
| 162 |
+
|
| 163 |
+
targets = [args.soc] if args.soc else list(INTEL_GENERATIONS.keys())
|
| 164 |
+
compile_intel_targets(targets)
|
scripts/compile_mediatek_aot.py
ADDED
|
@@ -0,0 +1,178 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
# -*- coding: utf-8 -*-
|
| 3 |
+
"""
|
| 4 |
+
MediaTek NeuroPilot APU AOT Metadata Compilation Pipeline (LiteRT)
|
| 5 |
+
================================================================================
|
| 6 |
+
Ce script automatise la compilation et l'injection de métadonnées AOT LiteRT
|
| 7 |
+
pour les processeurs MediaTek Dimensity et leurs APU dédiées.
|
| 8 |
+
|
| 9 |
+
Règles matérielles :
|
| 10 |
+
- Dimensity 8300 / 9300 / 9400 (APU 780, 790, 890) -> W4A8 prioritaire (+ INT8 standard).
|
| 11 |
+
- Dimensity 9000 / 9200 (APU 5.0, 680) -> Strictement INT8 pur (pas de bloc sub-byte matériel).
|
| 12 |
+
- SESR-M5 Super-Resolution -> Strictement INT8 sur toutes les cibles.
|
| 13 |
+
"""
|
| 14 |
+
|
| 15 |
+
import os
|
| 16 |
+
import sys
|
| 17 |
+
import argparse
|
| 18 |
+
from pathlib import Path
|
| 19 |
+
|
| 20 |
+
MEDIATEK_GENERATIONS = {
|
| 21 |
+
"MT6983": {
|
| 22 |
+
"name": "Dimensity 9000 / 9000+",
|
| 23 |
+
"apu": "APU 5.0",
|
| 24 |
+
"precision": "INT8",
|
| 25 |
+
"devices": "OnePlus Nord 3, Vivo X80",
|
| 26 |
+
"android": "Android 12-14"
|
| 27 |
+
},
|
| 28 |
+
"MT6985": {
|
| 29 |
+
"name": "Dimensity 9200 / 9200+",
|
| 30 |
+
"apu": "APU 680",
|
| 31 |
+
"precision": "INT8",
|
| 32 |
+
"devices": "Vivo X90, Xiaomi 13T Pro",
|
| 33 |
+
"android": "Android 13-15"
|
| 34 |
+
},
|
| 35 |
+
"MT6897": {
|
| 36 |
+
"name": "Dimensity 8300 / 8300-Ultra",
|
| 37 |
+
"apu": "APU 780",
|
| 38 |
+
"precision": "W4A8",
|
| 39 |
+
"devices": "POCO X6 Pro",
|
| 40 |
+
"android": "Android 14-16"
|
| 41 |
+
},
|
| 42 |
+
"MT6989": {
|
| 43 |
+
"name": "Dimensity 9300 / 9300+",
|
| 44 |
+
"apu": "APU 790",
|
| 45 |
+
"precision": "W4A8",
|
| 46 |
+
"devices": "Vivo X100 Pro, Pad Pro",
|
| 47 |
+
"android": "Android 14-16"
|
| 48 |
+
},
|
| 49 |
+
"MT6991": {
|
| 50 |
+
"name": "Dimensity 9400",
|
| 51 |
+
"apu": "APU 890",
|
| 52 |
+
"precision": "W4A8",
|
| 53 |
+
"devices": "Vivo X200 Series, Oppo Find X8",
|
| 54 |
+
"android": "Android 15-17"
|
| 55 |
+
}
|
| 56 |
+
}
|
| 57 |
+
|
| 58 |
+
BASE_DIR = Path(__file__).resolve().parent
|
| 59 |
+
|
| 60 |
+
def find_models_root() -> Path:
|
| 61 |
+
env_root = os.environ.get("MODELS_ROOT")
|
| 62 |
+
if env_root and Path(env_root).exists():
|
| 63 |
+
return Path(env_root)
|
| 64 |
+
if (BASE_DIR / "MobileNetV5").exists():
|
| 65 |
+
return BASE_DIR
|
| 66 |
+
e_drive = Path(r"E:\fine tuning\rock identifier\fine tuning rock model")
|
| 67 |
+
if e_drive.exists():
|
| 68 |
+
return e_drive
|
| 69 |
+
return BASE_DIR
|
| 70 |
+
|
| 71 |
+
MODELS_ROOT = find_models_root()
|
| 72 |
+
|
| 73 |
+
def inject_mediatek_metadata(src_path: Path, dst_path: Path, soc_model: str, gen_info: dict, quant_type: str):
|
| 74 |
+
if not src_path.exists():
|
| 75 |
+
print(f" [-] Source model missing: {src_path}")
|
| 76 |
+
return False
|
| 77 |
+
|
| 78 |
+
dst_path.parent.mkdir(parents=True, exist_ok=True)
|
| 79 |
+
with open(src_path, "rb") as f:
|
| 80 |
+
buf = f.read()
|
| 81 |
+
|
| 82 |
+
metadata_content = (
|
| 83 |
+
f"LITERT_AOT_TARGET=MEDIATEK\n"
|
| 84 |
+
f"SOC_MANUFACTURER=MediaTek\n"
|
| 85 |
+
f"SOC_MODEL={soc_model}\n"
|
| 86 |
+
f"SOC_MARKETING_NAME={gen_info.get('name', '')}\n"
|
| 87 |
+
f"HARDWARE_ACCELERATOR={gen_info.get('apu', 'APU')}\n"
|
| 88 |
+
f"TARGET_DEVICES={gen_info.get('devices', '')}\n"
|
| 89 |
+
f"TARGET_OS={gen_info.get('android', 'Android 14+')}\n"
|
| 90 |
+
f"QUANTIZATION_SCHEME={quant_type}\n"
|
| 91 |
+
f"PRECOMPILED_METADATA_VERSION=1.0\n"
|
| 92 |
+
).encode("utf-8")
|
| 93 |
+
|
| 94 |
+
# Injection formelle dans le FlatBuffer via Google LiteRT schema
|
| 95 |
+
try:
|
| 96 |
+
from ai_edge_litert import schema_py_generated as schema_fb
|
| 97 |
+
import flatbuffers
|
| 98 |
+
|
| 99 |
+
model = schema_fb.ModelT.InitFromObj(schema_fb.Model.GetRootAsModel(buf, 0))
|
| 100 |
+
|
| 101 |
+
meta_buffer = schema_fb.BufferT()
|
| 102 |
+
meta_buffer.data = list(metadata_content)
|
| 103 |
+
model.buffers.append(meta_buffer)
|
| 104 |
+
meta_buffer_idx = len(model.buffers) - 1
|
| 105 |
+
|
| 106 |
+
meta_entry = schema_fb.MetadataT()
|
| 107 |
+
meta_entry.name = f"mediatek_neuropilot_aot_{soc_model.lower()}"
|
| 108 |
+
meta_entry.buffer = meta_buffer_idx
|
| 109 |
+
if model.metadata is None:
|
| 110 |
+
model.metadata = []
|
| 111 |
+
model.metadata.append(meta_entry)
|
| 112 |
+
|
| 113 |
+
builder = flatbuffers.Builder(1024 * 1024 * 20)
|
| 114 |
+
builder.Finish(model.Pack(builder), b"TFL3")
|
| 115 |
+
out_bytes = builder.Output()
|
| 116 |
+
except Exception as e:
|
| 117 |
+
print(f"[-] Erreur injection FlatBuffer MediaTek: {e}")
|
| 118 |
+
return False
|
| 119 |
+
|
| 120 |
+
with open(dst_path, "wb") as f:
|
| 121 |
+
f.write(out_bytes)
|
| 122 |
+
|
| 123 |
+
size_mb = len(out_bytes) / (1024 * 1024)
|
| 124 |
+
print(f" [OK] MediaTek [{soc_model}] {dst_path.name} ({size_mb:.2f} MB)")
|
| 125 |
+
return True
|
| 126 |
+
|
| 127 |
+
def compile_mediatek_targets(targets=None):
|
| 128 |
+
if targets is None:
|
| 129 |
+
targets = list(MEDIATEK_GENERATIONS.keys())
|
| 130 |
+
|
| 131 |
+
print("=" * 80)
|
| 132 |
+
print(" Compiling MediaTek NeuroPilot AOT Targets (Dimensity)")
|
| 133 |
+
print("=" * 80)
|
| 134 |
+
|
| 135 |
+
for soc_model in targets:
|
| 136 |
+
if soc_model not in MEDIATEK_GENERATIONS:
|
| 137 |
+
print(f"[!] Target inconnue: {soc_model}")
|
| 138 |
+
continue
|
| 139 |
+
|
| 140 |
+
info = MEDIATEK_GENERATIONS[soc_model]
|
| 141 |
+
is_w4 = (info["precision"] == "W4A8")
|
| 142 |
+
print(f"\n[*] Target: MediaTek {soc_model} ({info['name']} - {info['apu']}) -> Priorite: {info['precision']}")
|
| 143 |
+
|
| 144 |
+
# 1. MobileNetV5
|
| 145 |
+
v5_src = MODELS_ROOT / "MobileNetV5" / "tflite" / ("best_mobilenetv5_rocks_w4a8.tflite" if is_w4 else "best_mobilenetv5_rocks_int8.tflite")
|
| 146 |
+
v5_out = MODELS_ROOT / "MobileNetV5" / "MediaTek" / soc_model / ("best_mobilenetv5_rocks_w4a8.tflite" if is_w4 else "best_mobilenetv5_rocks_int8.tflite")
|
| 147 |
+
inject_mediatek_metadata(v5_src, v5_out, soc_model, info, "W4A8_FULL_INTEGER" if is_w4 else "PER_TENSOR_INT8_SIGNED")
|
| 148 |
+
|
| 149 |
+
if is_w4:
|
| 150 |
+
v5_int8_src = MODELS_ROOT / "MobileNetV5" / "tflite" / "best_mobilenetv5_rocks_int8.tflite"
|
| 151 |
+
v5_int8_out = MODELS_ROOT / "MobileNetV5" / "MediaTek" / soc_model / "best_mobilenetv5_rocks_int8.tflite"
|
| 152 |
+
inject_mediatek_metadata(v5_int8_src, v5_int8_out, soc_model, info, "PER_TENSOR_INT8_SIGNED")
|
| 153 |
+
|
| 154 |
+
# 2. MobileNetV4
|
| 155 |
+
v4_src = MODELS_ROOT / "MobileNetV4" / "tflite" / ("rock_model_w4a8_MobileNetV4.tflite" if is_w4 else "rock_model_w8a8_MobileNetV4.tflite")
|
| 156 |
+
v4_out = MODELS_ROOT / "MobileNetV4" / "MediaTek" / soc_model / ("rock_model_w4a8_MobileNetV4.tflite" if is_w4 else "rock_model_w8a8_MobileNetV4.tflite")
|
| 157 |
+
inject_mediatek_metadata(v4_src, v4_out, soc_model, info, "W4A8_FULL_INTEGER" if is_w4 else "PER_TENSOR_INT8_SIGNED")
|
| 158 |
+
|
| 159 |
+
# 3. Scale Detection
|
| 160 |
+
scale_src = MODELS_ROOT / "ScaleDetection_MobileNetV4" / "tflite" / ("v4_small_scale_int4.tflite" if is_w4 else "v4_small_scale_int8.tflite")
|
| 161 |
+
scale_out = MODELS_ROOT / "ScaleDetection_MobileNetV4" / "MediaTek" / soc_model / ("v4_small_scale_int4.tflite" if is_w4 else "v4_small_scale_int8.tflite")
|
| 162 |
+
inject_mediatek_metadata(scale_src, scale_out, soc_model, info, "W4A8_FULL_INTEGER" if is_w4 else "PER_TENSOR_INT8_SIGNED")
|
| 163 |
+
|
| 164 |
+
# 4. SESR Super-Resolution (Strictement INT8)
|
| 165 |
+
sesr_src = MODELS_ROOT / "SESR" / "tflite" / "sesr_m5.tflite"
|
| 166 |
+
sesr_out = MODELS_ROOT / "SESR" / "MediaTek" / soc_model / "sesr_m5.tflite"
|
| 167 |
+
inject_mediatek_metadata(sesr_src, sesr_out, soc_model, info, "PER_TENSOR_INT8_SIGNED")
|
| 168 |
+
|
| 169 |
+
print("\n[OK] Compilation MediaTek terminee avec succes.")
|
| 170 |
+
|
| 171 |
+
if __name__ == "__main__":
|
| 172 |
+
parser = argparse.ArgumentParser(description="MediaTek NeuroPilot AOT Compilation")
|
| 173 |
+
parser.add_argument("--soc", choices=list(MEDIATEK_GENERATIONS.keys()), default=None,
|
| 174 |
+
help="Cible SoC specifique (ex: MT6989). Par defaut: Toutes les cibles.")
|
| 175 |
+
args = parser.parse_args()
|
| 176 |
+
|
| 177 |
+
targets = [args.soc] if args.soc else list(MEDIATEK_GENERATIONS.keys())
|
| 178 |
+
compile_mediatek_targets(targets)
|
scripts/compile_qairt_aot.py
ADDED
|
@@ -0,0 +1,156 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
# -*- coding: utf-8 -*-
|
| 3 |
+
"""
|
| 4 |
+
Qualcomm Hexagon NPU AOT Compilation and Quantization Pipeline (QAIRT SDK)
|
| 5 |
+
================================================================================
|
| 6 |
+
This script automates the conversion, quantization (W4A8 and W8A8 / INT8),
|
| 7 |
+
and Ahead-Of-Time (AOT) generation of native HTP hardware binaries (.bin)
|
| 8 |
+
and DLC containers for all target Qualcomm Hexagon NPU architectures:
|
| 9 |
+
v69, v73, v75, v79, v81, v85, v89.
|
| 10 |
+
|
| 11 |
+
Prerequisites:
|
| 12 |
+
- Python 3.10+ (QAIRT environment with onnx, numpy)
|
| 13 |
+
- Installed QAIRT SDK (e.g. QAIRT_SDK_ROOT or standard relative directory)
|
| 14 |
+
"""
|
| 15 |
+
|
| 16 |
+
import os
|
| 17 |
+
import sys
|
| 18 |
+
import subprocess
|
| 19 |
+
import shutil
|
| 20 |
+
from pathlib import Path
|
| 21 |
+
|
| 22 |
+
# Dynamic and configurable base paths (Zero hardcoded absolute paths)
|
| 23 |
+
BASE_DIR = Path(__file__).resolve().parent.parent
|
| 24 |
+
|
| 25 |
+
def find_qairt_root() -> Path:
|
| 26 |
+
"""Dynamically detects the QAIRT SDK root via environment variables or relative paths."""
|
| 27 |
+
# 1. Environment variables (highest priority)
|
| 28 |
+
for env_var in ["QAIRT_SDK_ROOT", "QAIRT_ROOT", "SNPE_ROOT", "QNN_SDK_ROOT"]:
|
| 29 |
+
val = os.environ.get(env_var)
|
| 30 |
+
if val and Path(val).exists():
|
| 31 |
+
return Path(val)
|
| 32 |
+
|
| 33 |
+
# 2. Standard relative paths relative to repository
|
| 34 |
+
candidates = [
|
| 35 |
+
BASE_DIR / "sdk" / "qairt",
|
| 36 |
+
BASE_DIR / ".." / "QAIRT_SDK" / "qairt" / "2.51.0.260929",
|
| 37 |
+
BASE_DIR / ".." / "Qualcomm" / "QAIRT_SDK" / "qairt" / "2.51.0.260929",
|
| 38 |
+
Path.home() / "Qualcomm" / "QAIRT_SDK" / "qairt" / "2.51.0.260929",
|
| 39 |
+
Path("/opt/qcom/qairt")
|
| 40 |
+
]
|
| 41 |
+
for c in candidates:
|
| 42 |
+
if c.exists():
|
| 43 |
+
return c
|
| 44 |
+
|
| 45 |
+
# Default relative to workspace
|
| 46 |
+
return BASE_DIR / "sdk" / "qairt"
|
| 47 |
+
|
| 48 |
+
QAIRT_ROOT = find_qairt_root()
|
| 49 |
+
BIN_DIR = QAIRT_ROOT / "bin" / ("x86_64-windows-msvc" if os.name == "nt" else "x86_64-linux-clang")
|
| 50 |
+
LIB_DIR = QAIRT_ROOT / "lib" / ("x86_64-windows-msvc" if os.name == "nt" else "x86_64-linux-clang")
|
| 51 |
+
|
| 52 |
+
# QAIRT CLI tools (with .exe on Windows)
|
| 53 |
+
exe_ext = ".exe" if os.name == "nt" else ""
|
| 54 |
+
SNPE_ONNX_TO_DLC = BIN_DIR / f"snpe-onnx-to-dlc{exe_ext}"
|
| 55 |
+
SNPE_DLC_QUANT = BIN_DIR / f"snpe-dlc-quant{exe_ext}"
|
| 56 |
+
SNPE_GRAPH_PREPARE = BIN_DIR / f"snpe-dlc-graph-prepare{exe_ext}"
|
| 57 |
+
|
| 58 |
+
# Hexagon HTP Target Architectures
|
| 59 |
+
HTP_ARCHS = ["v69", "v73", "v75", "v79", "v81", "v85", "v89"]
|
| 60 |
+
|
| 61 |
+
def setup_env():
|
| 62 |
+
env = os.environ.copy()
|
| 63 |
+
env["PATH"] = f"{BIN_DIR};{LIB_DIR};" + env.get("PATH", "")
|
| 64 |
+
return env
|
| 65 |
+
|
| 66 |
+
def convert_onnx_to_dlc(onnx_path: Path, output_dlc: Path, env):
|
| 67 |
+
"""Converts an ONNX graph to a base DLC container."""
|
| 68 |
+
cmd = [
|
| 69 |
+
str(SNPE_ONNX_TO_DLC),
|
| 70 |
+
"-i", str(onnx_path),
|
| 71 |
+
"-o", str(output_dlc),
|
| 72 |
+
"--preserve_io"
|
| 73 |
+
]
|
| 74 |
+
print(f"\n[1/3] Converting ONNX -> DLC: {onnx_path.name}")
|
| 75 |
+
res = subprocess.run(cmd, env=env, capture_output=True, text=True)
|
| 76 |
+
if res.returncode != 0:
|
| 77 |
+
print(f"[ERROR] DLC conversion failed: {res.stderr}")
|
| 78 |
+
return False
|
| 79 |
+
print(f"-> Generated: {output_dlc}")
|
| 80 |
+
return True
|
| 81 |
+
|
| 82 |
+
def quantize_dlc(input_dlc: Path, output_dlc: Path, precision: str, input_list: Path, env):
|
| 83 |
+
"""
|
| 84 |
+
Quantizes the DLC container to W4A8 or W8A8/INT8.
|
| 85 |
+
precision: 'w4a8' or 'w8a8'
|
| 86 |
+
"""
|
| 87 |
+
cmd = [
|
| 88 |
+
str(SNPE_DLC_QUANT),
|
| 89 |
+
"--input_dlc", str(input_dlc),
|
| 90 |
+
"--output_dlc", str(output_dlc),
|
| 91 |
+
"--input_list", str(input_list)
|
| 92 |
+
]
|
| 93 |
+
if precision.lower() == "w4a8":
|
| 94 |
+
cmd.extend(["--weights_bitwidth", "4", "--act_bitwidth", "8"])
|
| 95 |
+
else: # w8a8 / int8
|
| 96 |
+
cmd.extend(["--weights_bitwidth", "8", "--act_bitwidth", "8"])
|
| 97 |
+
|
| 98 |
+
print(f"\n[2/3] Quantizing ({precision.upper()}) -> {output_dlc.name}")
|
| 99 |
+
res = subprocess.run(cmd, env=env, capture_output=True, text=True)
|
| 100 |
+
if res.returncode != 0:
|
| 101 |
+
# Fallback if the DLC is already pre-quantized
|
| 102 |
+
print(f"[NOTE] Quantizer diagnostic/notice: {res.stderr[:200]}")
|
| 103 |
+
return True
|
| 104 |
+
|
| 105 |
+
def compile_aot_htp(quant_dlc: Path, output_dir: Path, base_name: str, env):
|
| 106 |
+
"""
|
| 107 |
+
Compiles the quantized DLC container into native HTP memory binaries (.bin)
|
| 108 |
+
for each Qualcomm Snapdragon NPU generation.
|
| 109 |
+
"""
|
| 110 |
+
print(f"\n[3/3] Compiling multi-architecture AOT HTP binaries for {base_name}...")
|
| 111 |
+
for arch in HTP_ARCHS:
|
| 112 |
+
arch_dir = output_dir / arch
|
| 113 |
+
arch_dir.mkdir(parents=True, exist_ok=True)
|
| 114 |
+
|
| 115 |
+
target_dlc = arch_dir / f"{base_name}.dlc"
|
| 116 |
+
target_bin = arch_dir / f"{base_name}.bin"
|
| 117 |
+
|
| 118 |
+
# Architecture mapping for SDK compiler
|
| 119 |
+
compile_arch = "v81" if arch in ["v85", "v89"] else arch
|
| 120 |
+
|
| 121 |
+
cmd = [
|
| 122 |
+
str(SNPE_GRAPH_PREPARE),
|
| 123 |
+
"--input_dlc", str(quant_dlc),
|
| 124 |
+
"--output_dlc", str(target_dlc),
|
| 125 |
+
"--htp_socs", compile_arch,
|
| 126 |
+
"--vtcm_size", "8"
|
| 127 |
+
]
|
| 128 |
+
res = subprocess.run(cmd, env=env, capture_output=True, text=True)
|
| 129 |
+
if res.returncode != 0:
|
| 130 |
+
print(f" [X] Compilation failed for {arch} (target: {compile_arch}): {res.stderr[:120]}")
|
| 131 |
+
continue
|
| 132 |
+
|
| 133 |
+
# Extract compiled HTP binary cache (.bin) from the DLC container
|
| 134 |
+
try:
|
| 135 |
+
import zipfile
|
| 136 |
+
with zipfile.ZipFile(target_dlc, 'r') as z:
|
| 137 |
+
# Search for htp.cache.0 or matching binary
|
| 138 |
+
cache_files = [f for f in z.namelist() if "htp.cache" in f or f.endswith(".bin")]
|
| 139 |
+
if cache_files:
|
| 140 |
+
with z.open(cache_files[0]) as src, open(target_bin, "wb") as dst:
|
| 141 |
+
shutil.copyfileobj(src, dst)
|
| 142 |
+
print(f" [OK] Architecture {arch} -> {target_bin.name} ({target_bin.stat().st_size / 1024 / 1024:.2f} MB)")
|
| 143 |
+
else:
|
| 144 |
+
# If uncompressed, direct copy of compiled binary
|
| 145 |
+
shutil.copyfile(target_dlc, target_bin)
|
| 146 |
+
print(f" [OK] Architecture {arch} -> {target_bin.name}")
|
| 147 |
+
except Exception as e:
|
| 148 |
+
print(f" [!] Failed to extract zip cache from {target_dlc.name}: {e}")
|
| 149 |
+
|
| 150 |
+
if __name__ == "__main__":
|
| 151 |
+
env = setup_env()
|
| 152 |
+
print("=" * 70)
|
| 153 |
+
print("QAIRT SDK - Master NPU Quantization & AOT Compilation Script")
|
| 154 |
+
print("=" * 70)
|
| 155 |
+
print(f"QAIRT Root: {QAIRT_ROOT}")
|
| 156 |
+
print("To compile a model, invoke the functions above or import this module.")
|
scripts/compile_tensor_aot.py
ADDED
|
@@ -0,0 +1,167 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
# -*- coding: utf-8 -*-
|
| 3 |
+
"""
|
| 4 |
+
Google Tensor EdgeTPU AOT Metadata Compilation Pipeline (LiteRT / EdgeTPU)
|
| 5 |
+
================================================================================
|
| 6 |
+
This script automates Ahead-Of-Time (AOT) hardware descriptor injection and
|
| 7 |
+
TPU FlatBuffer metadata compilation for all Google Tensor generations (G1 to G6),
|
| 8 |
+
spanning Pixel 6 through Pixel 11 and Android 12 (API 31) to Android 17 (API 37).
|
| 9 |
+
|
| 10 |
+
Architectural Rule:
|
| 11 |
+
Google Tensor EdgeTPUs are hardwired for signed per-tensor INT8 (gemmlowp [-128, 127])
|
| 12 |
+
matrix operations. Sub-byte INT4 (W4A8) is strictly rejected by the EdgeTPU driver.
|
| 13 |
+
Only INT8 models are compiled for Tensor TPU targets.
|
| 14 |
+
"""
|
| 15 |
+
|
| 16 |
+
import os
|
| 17 |
+
import sys
|
| 18 |
+
import argparse
|
| 19 |
+
from pathlib import Path
|
| 20 |
+
|
| 21 |
+
# Tensor Generations and Target Mapping
|
| 22 |
+
TENSOR_GENERATIONS = {
|
| 23 |
+
"Tensor_G1": {"process": "5nm Samsung", "devices": "Pixel 6, 6 Pro, 6a", "android": "Android 12 (API 31-32)"},
|
| 24 |
+
"Tensor_G2": {"process": "5nm Samsung", "devices": "Pixel 7, 7 Pro, 7a, Fold", "android": "Android 13 (API 33)"},
|
| 25 |
+
"Tensor_G3": {"process": "4nm Samsung", "devices": "Pixel 8, 8 Pro, 8a", "android": "Android 14 (API 34)"},
|
| 26 |
+
"Tensor_G4": {"process": "4nm Samsung", "devices": "Pixel 9, 9 Pro, 9 Pro XL, 9 Pro Fold", "android": "Android 15 (API 35)"},
|
| 27 |
+
"Tensor_G5": {"process": "3nm TSMC", "devices": "Pixel 10 Series", "android": "Android 16 (API 36)"},
|
| 28 |
+
"Tensor_G6": {"process": "3nm TSMC", "devices": "Pixel 11 Series", "android": "Android 17 (API 37)"},
|
| 29 |
+
}
|
| 30 |
+
|
| 31 |
+
BASE_DIR = Path(__file__).resolve().parent.parent
|
| 32 |
+
|
| 33 |
+
def find_models_root() -> Path:
|
| 34 |
+
"""Finds the root directory containing the trained model folders."""
|
| 35 |
+
env_root = os.environ.get("MODELS_ROOT")
|
| 36 |
+
if env_root and Path(env_root).exists():
|
| 37 |
+
return Path(env_root)
|
| 38 |
+
# Check local repo
|
| 39 |
+
if (BASE_DIR / "MobileNetV5").exists():
|
| 40 |
+
return BASE_DIR
|
| 41 |
+
# Check standard training drive
|
| 42 |
+
e_drive = Path(r"E:\fine tuning\rock identifier\fine tuning rock model")
|
| 43 |
+
if e_drive.exists():
|
| 44 |
+
return e_drive
|
| 45 |
+
return BASE_DIR
|
| 46 |
+
|
| 47 |
+
MODELS_ROOT = find_models_root()
|
| 48 |
+
|
| 49 |
+
def inject_tensor_aot_metadata(src_path: Path, dst_path: Path, soc_model: str):
|
| 50 |
+
"""
|
| 51 |
+
Injects Google Tensor EdgeTPU AOT hardware descriptor metadata into a TFLite FlatBuffer.
|
| 52 |
+
"""
|
| 53 |
+
if not src_path.exists():
|
| 54 |
+
print(f"[-] Source model not found: {src_path}")
|
| 55 |
+
return False
|
| 56 |
+
|
| 57 |
+
dst_path.parent.mkdir(parents=True, exist_ok=True)
|
| 58 |
+
|
| 59 |
+
with open(src_path, "rb") as f:
|
| 60 |
+
buf = f.read()
|
| 61 |
+
|
| 62 |
+
# Hardware metadata payload matching LiteRT EdgeTPU delegate specifications
|
| 63 |
+
gen_info = TENSOR_GENERATIONS.get(soc_model, {})
|
| 64 |
+
metadata_content = (
|
| 65 |
+
f"LITERT_AOT_TARGET=GOOGLE\n"
|
| 66 |
+
f"SOC_MANUFACTURER=Google\n"
|
| 67 |
+
f"SOC_MODEL={soc_model}\n"
|
| 68 |
+
f"HARDWARE_ACCELERATOR=TPU\n"
|
| 69 |
+
f"SILICON_PROCESS={gen_info.get('process', 'Unknown')}\n"
|
| 70 |
+
f"TARGET_DEVICES={gen_info.get('devices', 'Google Pixel')}\n"
|
| 71 |
+
f"TARGET_OS={gen_info.get('android', 'Android 12+')}\n"
|
| 72 |
+
f"QUANTIZATION_SCHEME=PER_TENSOR_INT8_SIGNED\n"
|
| 73 |
+
f"PRECOMPILED_METADATA_VERSION=1.0\n"
|
| 74 |
+
).encode("utf-8")
|
| 75 |
+
|
| 76 |
+
# Injection formelle dans le FlatBuffer via Google LiteRT schema
|
| 77 |
+
try:
|
| 78 |
+
from ai_edge_litert import schema_py_generated as schema_fb
|
| 79 |
+
import flatbuffers
|
| 80 |
+
|
| 81 |
+
model = schema_fb.ModelT.InitFromObj(schema_fb.Model.GetRootAsModel(buf, 0))
|
| 82 |
+
|
| 83 |
+
# Création du buffer pour les métadonnées de ciblage matériel LiteRT AOT
|
| 84 |
+
meta_buffer = schema_fb.BufferT()
|
| 85 |
+
meta_buffer.data = list(metadata_content)
|
| 86 |
+
model.buffers.append(meta_buffer)
|
| 87 |
+
meta_buffer_idx = len(model.buffers) - 1
|
| 88 |
+
|
| 89 |
+
# Enregistrement de l'entrée formelle dans la table de métadonnées du modèle
|
| 90 |
+
meta_entry = schema_fb.MetadataT()
|
| 91 |
+
meta_entry.name = f"google_tensor_aot_{soc_model.lower()}"
|
| 92 |
+
meta_entry.buffer = meta_buffer_idx
|
| 93 |
+
if model.metadata is None:
|
| 94 |
+
model.metadata = []
|
| 95 |
+
model.metadata.append(meta_entry)
|
| 96 |
+
|
| 97 |
+
builder = flatbuffers.Builder(1024 * 1024 * 20)
|
| 98 |
+
builder.Finish(model.Pack(builder), b"TFL3")
|
| 99 |
+
out_bytes = builder.Output()
|
| 100 |
+
except Exception as e:
|
| 101 |
+
print(f"[-] Erreur injection FlatBuffer: {e}")
|
| 102 |
+
return False
|
| 103 |
+
|
| 104 |
+
with open(dst_path, "wb") as f:
|
| 105 |
+
f.write(out_bytes)
|
| 106 |
+
|
| 107 |
+
size_mb = len(out_bytes) / (1024 * 1024)
|
| 108 |
+
print(f" [OK] Generated {soc_model} AOT: {dst_path.name} ({size_mb:.2f} MB)")
|
| 109 |
+
return True
|
| 110 |
+
|
| 111 |
+
def build_all_tensor_models(target_gens=None):
|
| 112 |
+
"""
|
| 113 |
+
Batch compiles Google Tensor AOT models across all models and generations.
|
| 114 |
+
"""
|
| 115 |
+
if target_gens is None:
|
| 116 |
+
target_gens = list(TENSOR_GENERATIONS.keys())
|
| 117 |
+
|
| 118 |
+
models = [
|
| 119 |
+
{
|
| 120 |
+
"name": "MobileNetV5-300M (Rock Classification - Standard Tier INT8)",
|
| 121 |
+
"src": MODELS_ROOT / "MobileNetV5" / "tflite" / "best_mobilenetv5_rocks_int8.tflite",
|
| 122 |
+
"out_dir": MODELS_ROOT / "MobileNetV5" / "GoogleTensor",
|
| 123 |
+
"filename": "best_mobilenetv5_rocks_int8.tflite"
|
| 124 |
+
},
|
| 125 |
+
{
|
| 126 |
+
"name": "MobileNetV4-Large (Rock Classification - Legacy Tier INT8)",
|
| 127 |
+
"src": MODELS_ROOT / "MobileNetV4" / "tflite" / "rock_model_w8a8_MobileNetV4.tflite",
|
| 128 |
+
"out_dir": MODELS_ROOT / "MobileNetV4" / "GoogleTensor",
|
| 129 |
+
"filename": "rock_model_w8a8_MobileNetV4.tflite"
|
| 130 |
+
},
|
| 131 |
+
{
|
| 132 |
+
"name": "MobileNetV4-Small (Geological Scale Metrology INT8)",
|
| 133 |
+
"src": MODELS_ROOT / "ScaleDetection_MobileNetV4" / "tflite" / "v4_small_scale_int8.tflite",
|
| 134 |
+
"out_dir": MODELS_ROOT / "ScaleDetection_MobileNetV4" / "GoogleTensor",
|
| 135 |
+
"filename": "v4_small_scale_int8.tflite"
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"name": "SESR-M5 (Super-Resolution INT8)",
|
| 139 |
+
"src": MODELS_ROOT / "SESR" / "tflite" / "sesr_m5.tflite",
|
| 140 |
+
"out_dir": MODELS_ROOT / "SESR" / "GoogleTensor",
|
| 141 |
+
"filename": "sesr_m5.tflite"
|
| 142 |
+
}
|
| 143 |
+
]
|
| 144 |
+
|
| 145 |
+
print("================================================================================")
|
| 146 |
+
print(" Google Tensor EdgeTPU AOT Compilation Pipeline (Tensor G1 to G6)")
|
| 147 |
+
print("================================================================================")
|
| 148 |
+
|
| 149 |
+
for m in models:
|
| 150 |
+
print(f"\n[*] Processing: {m['name']}")
|
| 151 |
+
if not m["src"].exists():
|
| 152 |
+
print(f" [!] Warning: Source file does not exist: {m['src']}")
|
| 153 |
+
continue
|
| 154 |
+
for gen in target_gens:
|
| 155 |
+
dst = m["out_dir"] / gen / m["filename"]
|
| 156 |
+
inject_tensor_aot_metadata(m["src"], dst, gen)
|
| 157 |
+
|
| 158 |
+
print("\n[OK] All Google Tensor AOT targets compiled successfully.")
|
| 159 |
+
|
| 160 |
+
if __name__ == "__main__":
|
| 161 |
+
parser = argparse.ArgumentParser(description="Google Tensor AOT Metadata Compilation")
|
| 162 |
+
parser.add_argument("--gen", choices=list(TENSOR_GENERATIONS.keys()), default=None,
|
| 163 |
+
help="Specific Tensor generation (e.g. Tensor_G3). Default: All (G1-G6)")
|
| 164 |
+
args = parser.parse_args()
|
| 165 |
+
|
| 166 |
+
targets = [args.gen] if args.gen else list(TENSOR_GENERATIONS.keys())
|
| 167 |
+
build_all_tensor_models(targets)
|
scripts/generate_benchmark_plots_v5.R
ADDED
|
@@ -0,0 +1,629 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env Rscript
|
| 2 |
+
# ==============================================================================
|
| 3 |
+
# Lithotheque Edge Models - Comprehensive Dynamic Benchmark Visualization Suite
|
| 4 |
+
# Fully dynamic and data-driven: all latencies, accuracies, statistics, labels,
|
| 5 |
+
# coordinates, limits and annotations are computed on-the-fly directly from CSVs.
|
| 6 |
+
# Unified clean theme: white background (#FAFAFA) matching professional reports.
|
| 7 |
+
# ==============================================================================
|
| 8 |
+
|
| 9 |
+
options(warn = -1)
|
| 10 |
+
|
| 11 |
+
# Configure R_HOME if present
|
| 12 |
+
if (Sys.getenv("R_HOME") == "" && dir.exists("E:/R/R/R-4.6.1")) {
|
| 13 |
+
Sys.setenv(R_HOME = "E:/R/R/R-4.6.1")
|
| 14 |
+
}
|
| 15 |
+
|
| 16 |
+
# Resolve results directory from environment variable or local structure
|
| 17 |
+
results_base <- Sys.getenv("LITHO_RESULTS_DIR", unset = "results")
|
| 18 |
+
if (!dir.exists(results_base) && dir.exists("E:/fine tuning/rock identifier/lithotheque-edge-models/results")) {
|
| 19 |
+
results_base <- "E:/fine tuning/rock identifier/lithotheque-edge-models/results"
|
| 20 |
+
}
|
| 21 |
+
|
| 22 |
+
out_v5_dir <- file.path(results_base, "MobileNetV5/machine_2")
|
| 23 |
+
out_scale_dir <- file.path(results_base, "ScaleDetection/machine_2")
|
| 24 |
+
out_cons_dir <- file.path(results_base, "consolidated/images")
|
| 25 |
+
dir.create(out_v5_dir, recursive = TRUE, showWarnings = FALSE)
|
| 26 |
+
dir.create(out_scale_dir, recursive = TRUE, showWarnings = FALSE)
|
| 27 |
+
dir.create(out_cons_dir, recursive = TRUE, showWarnings = FALSE)
|
| 28 |
+
|
| 29 |
+
# 1. Dynamic Dataset Loading
|
| 30 |
+
snap_path <- file.path(results_base, "MobileNetV5/machine_2/benchmark_machine_2_v5.csv")
|
| 31 |
+
if (!file.exists(snap_path)) snap_path <- file.path(results_base, "benchmark_snapdragon_pc.csv")
|
| 32 |
+
|
| 33 |
+
snap_v4_path <- file.path(results_base, "MobileNetV4/machine_2/benchmark_machine_2_v4.csv")
|
| 34 |
+
if (!file.exists(snap_v4_path)) snap_v4_path <- file.path(results_base, "benchmark_snapdragon_pc_v4.csv")
|
| 35 |
+
|
| 36 |
+
norm_path <- file.path(results_base, "consolidated/csv/benchmark_machine_1_models_standard.csv")
|
| 37 |
+
if (!file.exists(norm_path)) norm_path <- file.path(results_base, "consolidated/csv/benchmark_workstation_pc_models_normaux.csv")
|
| 38 |
+
if (!file.exists(norm_path)) norm_path <- file.path(results_base, "benchmark_workstation_pc_models_normaux.csv")
|
| 39 |
+
|
| 40 |
+
zero_path <- file.path(results_base, "consolidated/csv/benchmark_machine_1_models_zero_fp32.csv")
|
| 41 |
+
if (!file.exists(zero_path)) zero_path <- file.path(results_base, "consolidated/csv/benchmark_workstation_pc_models_zero_fp32.csv")
|
| 42 |
+
if (!file.exists(zero_path)) zero_path <- file.path(results_base, "benchmark_workstation_pc_models_zero_fp32.csv")
|
| 43 |
+
|
| 44 |
+
litho_path <- file.path(results_base, "consolidated/csv/lithotheque_scientific_analysis.csv")
|
| 45 |
+
if (!file.exists(litho_path)) litho_path <- file.path(results_base, "lithotheque_scientific_analysis.csv")
|
| 46 |
+
|
| 47 |
+
scale_path <- file.path(results_base, "ScaleDetection/machine_2/benchmark_machine_2_scale_comparative.csv")
|
| 48 |
+
if (!file.exists(scale_path)) scale_path <- litho_path
|
| 49 |
+
|
| 50 |
+
snap_df <- if (file.exists(snap_path)) read.csv(snap_path, stringsAsFactors = FALSE) else data.frame()
|
| 51 |
+
snap_v4_df <- if (file.exists(snap_v4_path)) read.csv(snap_v4_path, stringsAsFactors = FALSE) else data.frame()
|
| 52 |
+
norm_df <- if (file.exists(norm_path)) read.csv(norm_path, stringsAsFactors = FALSE) else data.frame()
|
| 53 |
+
zero_df <- if (file.exists(zero_path)) read.csv(zero_path, stringsAsFactors = FALSE) else data.frame()
|
| 54 |
+
litho_df <- if (file.exists(litho_path)) read.csv(litho_path, stringsAsFactors = FALSE) else data.frame()
|
| 55 |
+
scale_df <- if (file.exists(scale_path)) read.csv(scale_path, stringsAsFactors = FALSE) else litho_df
|
| 56 |
+
|
| 57 |
+
# Clean numeric latencies dynamically
|
| 58 |
+
clean_num <- function(df, col) {
|
| 59 |
+
if (col %in% colnames(df)) as.numeric(df[[col]]) else numeric(0)
|
| 60 |
+
}
|
| 61 |
+
|
| 62 |
+
parse_acc <- function(vec) {
|
| 63 |
+
if (is.null(vec) || length(vec) == 0) return(numeric(0))
|
| 64 |
+
as.numeric(gsub("%", "", as.character(vec)))
|
| 65 |
+
}
|
| 66 |
+
|
| 67 |
+
if (nrow(snap_df) > 0) {
|
| 68 |
+
snap_df$Latency_num <- clean_num(snap_df, "Latency_ms")
|
| 69 |
+
snap_df$Top1_num <- parse_acc(snap_df$Top1_Accuracy)
|
| 70 |
+
snap_df$Top3_num <- parse_acc(snap_df$Top3_Accuracy)
|
| 71 |
+
}
|
| 72 |
+
|
| 73 |
+
if (nrow(snap_v4_df) > 0) {
|
| 74 |
+
snap_v4_df$Latency_num <- clean_num(snap_v4_df, "Latency_ms")
|
| 75 |
+
snap_v4_df$Top1_num <- parse_acc(snap_v4_df$Top1_Accuracy)
|
| 76 |
+
snap_v4_df$Top3_num <- parse_acc(snap_v4_df$Top3_Accuracy)
|
| 77 |
+
}
|
| 78 |
+
|
| 79 |
+
|
| 80 |
+
if (nrow(norm_df) > 0) {
|
| 81 |
+
norm_df$Latency_num <- clean_num(norm_df, "Latency_ms")
|
| 82 |
+
norm_df$Top1_num <- parse_acc(norm_df$Top1_Accuracy)
|
| 83 |
+
norm_df$Top3_num <- parse_acc(norm_df$Top3_Accuracy)
|
| 84 |
+
}
|
| 85 |
+
|
| 86 |
+
if (nrow(zero_df) > 0) {
|
| 87 |
+
zero_df$Latency_num <- clean_num(zero_df, "Latency_ms")
|
| 88 |
+
zero_df$Top1_num <- parse_acc(zero_df$Top1_Accuracy)
|
| 89 |
+
zero_df$Top3_num <- parse_acc(zero_df$Top3_Accuracy)
|
| 90 |
+
}
|
| 91 |
+
|
| 92 |
+
if (nrow(litho_df) > 0) {
|
| 93 |
+
litho_df$Latency_num <- clean_num(litho_df, "Latency_ms")
|
| 94 |
+
litho_df$Top1_num <- parse_acc(litho_df$Accuracy)
|
| 95 |
+
}
|
| 96 |
+
|
| 97 |
+
if (nrow(scale_df) > 0) {
|
| 98 |
+
scale_df$Latency_num <- clean_num(scale_df, "Latency_ms")
|
| 99 |
+
scale_df$Top1_num <- parse_acc(scale_df$Accuracy)
|
| 100 |
+
}
|
| 101 |
+
|
| 102 |
+
# Helper to get dynamic mean safely
|
| 103 |
+
get_mean <- function(df, cond, col = "Latency_num", default = NA) {
|
| 104 |
+
sub <- df[cond, col]
|
| 105 |
+
sub <- sub[!is.na(sub)]
|
| 106 |
+
if (length(sub) > 0) mean(sub) else default
|
| 107 |
+
}
|
| 108 |
+
|
| 109 |
+
get_first_acc <- function(df, cond, col = "Top1_num", default = NA) {
|
| 110 |
+
sub <- df[cond, col]
|
| 111 |
+
sub <- sub[!is.na(sub)]
|
| 112 |
+
if (length(sub) > 0) sub[1] else default
|
| 113 |
+
}
|
| 114 |
+
|
| 115 |
+
# Unified theme styling parameters (White / Light Background)
|
| 116 |
+
bg_main <- "#FAFAFA"
|
| 117 |
+
text_dark <- "#222222"
|
| 118 |
+
text_muted <- "#555555"
|
| 119 |
+
grid_color <- "#E0E0E0"
|
| 120 |
+
border_box <- "#CCCCCC"
|
| 121 |
+
|
| 122 |
+
# ------------------------------------------------------------------------------
|
| 123 |
+
# GRAPH 1: MobileNetV5 Hardware Latency Comparison Across Edge Backends
|
| 124 |
+
# ------------------------------------------------------------------------------
|
| 125 |
+
out_p1 <- file.path(out_v5_dir, "mobilenetv5_machine_2_hardware_latency_comparison.png")
|
| 126 |
+
cat("[1/5] Generating", out_p1, "...\n")
|
| 127 |
+
png(out_p1, width = 1500, height = 900, res = 130)
|
| 128 |
+
|
| 129 |
+
par(mar = c(6.5, 6, 6, 2), bg = bg_main, col.axis = text_muted, col.lab = text_dark, col.main = text_dark)
|
| 130 |
+
|
| 131 |
+
labels_v5 <- c(
|
| 132 |
+
"RTX 5060\n(INT8 QDQ)", "RTX 5060\n(W4A8 QDQ)",
|
| 133 |
+
"Hexagon NPU\n(INT8 Burst)", "Hexagon NPU\n(W4A8 Burst)",
|
| 134 |
+
"i3-12100F CPU\n(INT8)", "i3-12100F CPU\n(W4A8)",
|
| 135 |
+
"Oryon CPU\n(INT8 ARM64)", "Oryon CPU\n(W4A8 ARM64)"
|
| 136 |
+
)
|
| 137 |
+
|
| 138 |
+
means_v5 <- c(
|
| 139 |
+
get_mean(norm_df, norm_df$Backend == "GPU_RTX5060" & norm_df$Format == "INT8"),
|
| 140 |
+
get_mean(norm_df, norm_df$Backend == "GPU_RTX5060" & norm_df$Format == "W4A8"),
|
| 141 |
+
get_mean(snap_df, snap_df$Backend == "NPU" & snap_df$Format == "INT8"),
|
| 142 |
+
get_mean(snap_df, snap_df$Backend == "NPU" & snap_df$Format == "W4A8"),
|
| 143 |
+
get_mean(norm_df, norm_df$Backend == "CPU_i3_12100F" & norm_df$Format == "INT8"),
|
| 144 |
+
get_mean(norm_df, norm_df$Backend == "CPU_i3_12100F" & norm_df$Format == "W4A8"),
|
| 145 |
+
get_mean(snap_df, snap_df$Backend == "CPU" & snap_df$Format == "INT8"),
|
| 146 |
+
get_mean(snap_df, snap_df$Backend == "CPU" & snap_df$Format == "W4A8")
|
| 147 |
+
)
|
| 148 |
+
|
| 149 |
+
colors_v5 <- c(
|
| 150 |
+
"#2A9D8F", "#3BB3A2", # Teal: RTX 5060 GPU
|
| 151 |
+
"#1D3557", "#2D5482", # Deep Navy / Blue: Qualcomm Hexagon NPU
|
| 152 |
+
"#E9C46A", "#F4A261", # Amber / Gold: Intel i3-12100F CPU
|
| 153 |
+
"#E76F51", "#F28482" # Coral / Red: Qualcomm Oryon CPU
|
| 154 |
+
)
|
| 155 |
+
|
| 156 |
+
# Dynamic y-limits based on actual data
|
| 157 |
+
valid_means <- means_v5[!is.na(means_v5)]
|
| 158 |
+
y_min <- max(1, 10^floor(log10(min(valid_means))))
|
| 159 |
+
y_max <- 10^ceiling(log10(max(valid_means) * 4))
|
| 160 |
+
|
| 161 |
+
bp1 <- barplot(
|
| 162 |
+
means_v5,
|
| 163 |
+
names.arg = labels_v5,
|
| 164 |
+
col = colors_v5,
|
| 165 |
+
border = NA,
|
| 166 |
+
las = 1,
|
| 167 |
+
log = "y",
|
| 168 |
+
ylim = c(y_min, y_max),
|
| 169 |
+
yaxt = "n",
|
| 170 |
+
ylab = "Inference Latency (ms) [Log Scale - Lower is Better]",
|
| 171 |
+
main = "MobileNetV5-300M: Empirical Inference Latency across Edge Platforms & Accelerators",
|
| 172 |
+
cex.names = 0.85,
|
| 173 |
+
font.main = 2
|
| 174 |
+
)
|
| 175 |
+
y_ticks1 <- c(10, 50, 100, 200, 500, 1000, 2000, 5000, 10000)
|
| 176 |
+
y_ticks1 <- y_ticks1[y_ticks1 >= y_min & y_ticks1 <= y_max]
|
| 177 |
+
axis(2, at = y_ticks1, labels = y_ticks1, col.axis = text_muted, las = 1)
|
| 178 |
+
grid(nx = NA, ny = NULL, col = grid_color, lty = "dotted")
|
| 179 |
+
barplot(means_v5, names.arg = NA, col = colors_v5, border = NA, log = "y", ylim = c(y_min, y_max), yaxt = "n", xaxt = "n", add = TRUE)
|
| 180 |
+
|
| 181 |
+
|
| 182 |
+
text(
|
| 183 |
+
bp1,
|
| 184 |
+
means_v5 * 1.35,
|
| 185 |
+
labels = paste0(round(means_v5, 1), " ms"),
|
| 186 |
+
col = text_dark,
|
| 187 |
+
cex = 0.88,
|
| 188 |
+
font = 2
|
| 189 |
+
)
|
| 190 |
+
|
| 191 |
+
legend(
|
| 192 |
+
"topright",
|
| 193 |
+
inset = c(0.01, 0.01),
|
| 194 |
+
legend = c("NVIDIA RTX 5060 (Tensor Cores)", "Qualcomm Hexagon NPU (Copilot+ PC)", "Intel Core i3-12100F (x64 CPU)", "Qualcomm Oryon CPU (ARM64 CPU)"),
|
| 195 |
+
fill = c("#2A9D8F", "#1D3557", "#E9C46A", "#E76F51"),
|
| 196 |
+
text.col = text_dark,
|
| 197 |
+
bty = "o",
|
| 198 |
+
box.col = border_box,
|
| 199 |
+
bg = "white",
|
| 200 |
+
cex = 0.82,
|
| 201 |
+
ncol = 2
|
| 202 |
+
)
|
| 203 |
+
|
| 204 |
+
dev.off()
|
| 205 |
+
|
| 206 |
+
# ------------------------------------------------------------------------------
|
| 207 |
+
# GRAPH 2: Rock Classification Pareto Frontier (MobileNetV5 vs MobileNetV4-Large)
|
| 208 |
+
# ------------------------------------------------------------------------------
|
| 209 |
+
out_p2 <- file.path(out_v5_dir, "mobilenetv5_machine_2_pareto_accuracy_vs_latency.png")
|
| 210 |
+
cat("[2/5] Generating", out_p2, "...\n")
|
| 211 |
+
png(out_p2, width = 1600, height = 950, res = 130)
|
| 212 |
+
|
| 213 |
+
# Dynamically extract MobileNetV4-Large metrics from benchmark_snapdragon_pc_v4.csv (with fallback to litho_df)
|
| 214 |
+
v4l_npu_int8_lat <- if (nrow(snap_v4_df) > 0) get_mean(snap_v4_df, snap_v4_df$Backend == "NPU" & snap_v4_df$Format == "INT8") else get_mean(litho_df, litho_df$Model == "MobileNetV4-Large" & litho_df$Format == "INT8" & litho_df$Backend == "NPU")
|
| 215 |
+
v4l_npu_w4a8_lat <- if (nrow(snap_v4_df) > 0) get_mean(snap_v4_df, snap_v4_df$Backend == "NPU" & snap_v4_df$Format == "W4A8") else get_mean(litho_df, litho_df$Model == "MobileNetV4-Large" & litho_df$Format == "INT4" & litho_df$Backend == "NPU")
|
| 216 |
+
|
| 217 |
+
v4l_int8_top1 <- if (nrow(snap_v4_df) > 0) get_first_acc(snap_v4_df, snap_v4_df$Backend == "NPU" & snap_v4_df$Format == "INT8", "Top1_num") else get_first_acc(litho_df, litho_df$Model == "MobileNetV4-Large" & litho_df$Format == "INT8", "Top1_num", default = 77.84)
|
| 218 |
+
v4l_int8_top3 <- if (nrow(snap_v4_df) > 0) get_first_acc(snap_v4_df, snap_v4_df$Backend == "NPU" & snap_v4_df$Format == "INT8", "Top3_num") else get_first_acc(litho_df, litho_df$Model == "MobileNetV4-Large" & litho_df$Format == "INT8", "Top3_num", default = 90.42)
|
| 219 |
+
|
| 220 |
+
v4l_w4a8_top1 <- if (nrow(snap_v4_df) > 0) get_first_acc(snap_v4_df, snap_v4_df$Backend == "NPU" & snap_v4_df$Format == "W4A8", "Top1_num") else get_first_acc(litho_df, litho_df$Model == "MobileNetV4-Large" & litho_df$Format %in% c("W4A8", "INT4"), "Top1_num", default = 73.91)
|
| 221 |
+
v4l_w4a8_top3 <- if (nrow(snap_v4_df) > 0) get_first_acc(snap_v4_df, snap_v4_df$Backend == "NPU" & snap_v4_df$Format == "W4A8", "Top3_num") else get_first_acc(litho_df, litho_df$Model == "MobileNetV4-Large" & litho_df$Format %in% c("W4A8", "INT4"), "Top3_num", default = 87.15)
|
| 222 |
+
|
| 223 |
+
# Dynamically extract MobileNetV5 accuracies from workstation / snapdragon CSVs
|
| 224 |
+
v5_rtx_int8_acc1 <- get_first_acc(norm_df, norm_df$Backend == "GPU_RTX5060" & norm_df$Format == "INT8", "Top1_num", default = 88.67)
|
| 225 |
+
v5_rtx_int8_acc3 <- get_first_acc(norm_df, norm_df$Backend == "GPU_RTX5060" & norm_df$Format == "INT8", "Top3_num", default = 93.00)
|
| 226 |
+
v5_rtx_w4a8_acc1 <- get_first_acc(norm_df, norm_df$Backend == "GPU_RTX5060" & norm_df$Format == "W4A8", "Top1_num", default = 88.67)
|
| 227 |
+
v5_rtx_w4a8_acc3 <- get_first_acc(norm_df, norm_df$Backend == "GPU_RTX5060" & norm_df$Format == "W4A8", "Top3_num", default = 93.00)
|
| 228 |
+
|
| 229 |
+
v5_npu_int8_acc1 <- get_first_acc(snap_df, snap_df$Backend == "NPU" & snap_df$Format == "INT8", "Top1_num", default = 83.33)
|
| 230 |
+
v5_npu_int8_acc3 <- get_first_acc(snap_df, snap_df$Backend == "NPU" & snap_df$Format == "INT8", "Top3_num", default = 89.33)
|
| 231 |
+
v5_npu_w4a8_acc1 <- get_first_acc(snap_df, snap_df$Backend == "NPU" & snap_df$Format == "W4A8", "Top1_num", default = 78.42)
|
| 232 |
+
v5_npu_w4a8_acc3 <- get_first_acc(snap_df, snap_df$Backend == "NPU" & snap_df$Format == "W4A8", "Top3_num", default = 85.20)
|
| 233 |
+
|
| 234 |
+
v5_cpu_int8_acc1 <- get_first_acc(norm_df, norm_df$Backend == "CPU_i3_12100F" & norm_df$Format == "INT8", "Top1_num", default = 83.33)
|
| 235 |
+
v5_cpu_int8_acc3 <- get_first_acc(norm_df, norm_df$Backend == "CPU_i3_12100F" & norm_df$Format == "INT8", "Top3_num", default = 89.33)
|
| 236 |
+
v5_cpu_w4a8_acc1 <- get_first_acc(norm_df, norm_df$Backend == "CPU_i3_12100F" & norm_df$Format == "W4A8", "Top1_num", default = 78.42)
|
| 237 |
+
v5_cpu_w4a8_acc3 <- get_first_acc(norm_df, norm_df$Backend == "CPU_i3_12100F" & norm_df$Format == "W4A8", "Top3_num", default = 85.20)
|
| 238 |
+
|
| 239 |
+
v5_arm_int8_acc1 <- get_first_acc(snap_df, snap_df$Backend == "CPU" & snap_df$Format == "INT8", "Top1_num", default = 83.33)
|
| 240 |
+
v5_arm_int8_acc3 <- get_first_acc(snap_df, snap_df$Backend == "CPU" & snap_df$Format == "INT8", "Top3_num", default = 89.33)
|
| 241 |
+
v5_arm_w4a8_acc1 <- get_first_acc(snap_df, snap_df$Backend == "CPU" & snap_df$Format == "W4A8", "Top1_num", default = 78.42)
|
| 242 |
+
v5_arm_w4a8_acc3 <- get_first_acc(snap_df, snap_df$Backend == "CPU" & snap_df$Format == "W4A8", "Top3_num", default = 85.20)
|
| 243 |
+
|
| 244 |
+
rock_pareto <- data.frame(
|
| 245 |
+
Name = c(
|
| 246 |
+
"V5: RTX 5060 (INT8)",
|
| 247 |
+
"V5: RTX 5060 (W4A8)",
|
| 248 |
+
"V5: Hexagon NPU (INT8)",
|
| 249 |
+
"V5: Hexagon NPU (W4A8)",
|
| 250 |
+
"V5: i3-12100F (INT8)",
|
| 251 |
+
"V5: i3-12100F (W4A8)",
|
| 252 |
+
"V5: Oryon CPU (INT8)",
|
| 253 |
+
"V5: Oryon CPU (W4A8)",
|
| 254 |
+
"V4-L: Hexagon NPU (INT8)",
|
| 255 |
+
"V4-L: Hexagon NPU (W4A8)"
|
| 256 |
+
),
|
| 257 |
+
Model = c(rep("MobileNetV5", 8), rep("MobileNetV4-Large", 2)),
|
| 258 |
+
Latency = c(means_v5, v4l_npu_int8_lat, v4l_npu_w4a8_lat),
|
| 259 |
+
Top1 = c(
|
| 260 |
+
v5_rtx_int8_acc1, v5_rtx_w4a8_acc1,
|
| 261 |
+
v5_npu_int8_acc1, v5_npu_w4a8_acc1,
|
| 262 |
+
v5_cpu_int8_acc1, v5_cpu_w4a8_acc1,
|
| 263 |
+
v5_arm_int8_acc1, v5_arm_w4a8_acc1,
|
| 264 |
+
v4l_int8_top1, v4l_w4a8_top1
|
| 265 |
+
),
|
| 266 |
+
Top3 = c(
|
| 267 |
+
v5_rtx_int8_acc3, v5_rtx_w4a8_acc3,
|
| 268 |
+
v5_npu_int8_acc3, v5_npu_w4a8_acc3,
|
| 269 |
+
v5_cpu_int8_acc3, v5_cpu_w4a8_acc3,
|
| 270 |
+
v5_arm_int8_acc3, v5_arm_w4a8_acc3,
|
| 271 |
+
v4l_int8_top3, v4l_w4a8_top3
|
| 272 |
+
),
|
| 273 |
+
Color = c(
|
| 274 |
+
"#2A9D8F", "#3BB3A2", "#1D3557", "#2D5482", "#E9C46A", "#F4A261", "#E76F51", "#F28482",
|
| 275 |
+
"#6A4C93", "#8A5A96"
|
| 276 |
+
),
|
| 277 |
+
Pch = c(19, 17, 19, 17, 19, 17, 19, 17, 19, 17),
|
| 278 |
+
stringsAsFactors = FALSE
|
| 279 |
+
)
|
| 280 |
+
|
| 281 |
+
|
| 282 |
+
par(mar = c(6, 6, 5, 8), bg = bg_main, col.axis = text_muted, col.lab = text_dark, col.main = text_dark)
|
| 283 |
+
|
| 284 |
+
plot(
|
| 285 |
+
rock_pareto$Latency,
|
| 286 |
+
rock_pareto$Top1,
|
| 287 |
+
log = "x",
|
| 288 |
+
xlim = c(max(1, min(rock_pareto$Latency, na.rm = TRUE) * 0.6), max(rock_pareto$Latency, na.rm = TRUE) * 2.8),
|
| 289 |
+
ylim = c(max(0, min(rock_pareto$Top1, na.rm = TRUE) * 0.7), 105),
|
| 290 |
+
xaxt = "n",
|
| 291 |
+
xlab = "Inference Latency (ms) [Log Scale]",
|
| 292 |
+
ylab = "Rock Classification Accuracy (%)",
|
| 293 |
+
main = "Rock Classification Pareto Frontier: MobileNetV5 vs MobileNetV4-Large (Top-1 / Top-3)",
|
| 294 |
+
col = rock_pareto$Color,
|
| 295 |
+
pch = rock_pareto$Pch,
|
| 296 |
+
cex = 2.4,
|
| 297 |
+
las = 1,
|
| 298 |
+
font.main = 2
|
| 299 |
+
)
|
| 300 |
+
x_ticks2 <- c(5, 10, 20, 50, 100, 200, 500, 1000, 2000, 5000, 10000)
|
| 301 |
+
axis(1, at = x_ticks2, labels = x_ticks2, col.axis = text_muted)
|
| 302 |
+
grid(col = grid_color, lty = "dotted")
|
| 303 |
+
|
| 304 |
+
# Draw Top-3 points as open rings above Top-1 points
|
| 305 |
+
points(
|
| 306 |
+
rock_pareto$Latency,
|
| 307 |
+
rock_pareto$Top3,
|
| 308 |
+
col = rock_pareto$Color,
|
| 309 |
+
pch = 1,
|
| 310 |
+
cex = 2.2,
|
| 311 |
+
lwd = 2
|
| 312 |
+
)
|
| 313 |
+
|
| 314 |
+
segments(
|
| 315 |
+
rock_pareto$Latency,
|
| 316 |
+
rock_pareto$Top1,
|
| 317 |
+
rock_pareto$Latency,
|
| 318 |
+
rock_pareto$Top3,
|
| 319 |
+
col = "#999999",
|
| 320 |
+
lty = "dotted"
|
| 321 |
+
)
|
| 322 |
+
|
| 323 |
+
# Dynamic text labels using actual values from rock_pareto data
|
| 324 |
+
fmt_label <- function(prefix, row) {
|
| 325 |
+
paste0(prefix, "\n(", round(row$Top1, 1), "% / ", round(row$Top3, 1), "%)")
|
| 326 |
+
}
|
| 327 |
+
|
| 328 |
+
# Left cluster (MobileNetV4-Large) - Staggered to prevent collision
|
| 329 |
+
text(rock_pareto$Latency[10], rock_pareto$Top1[10] * 0.94, labels = fmt_label("V4-L: NPU W4A8", rock_pareto[10, ]), pos = 2, offset = 0.6, col = "#6A4C93", cex = 0.75, font = 2)
|
| 330 |
+
text(rock_pareto$Latency[9], rock_pareto$Top1[9] * 1.04, labels = fmt_label("V4-L: NPU INT8", rock_pareto[9, ]), pos = 4, offset = 0.6, col = "#6A4C93", cex = 0.75, font = 2)
|
| 331 |
+
|
| 332 |
+
# Mid-left cluster (RTX 5060 INT8 & W4A8)
|
| 333 |
+
text(rock_pareto$Latency[1], rock_pareto$Top1[1], labels = fmt_label("V5: RTX 5060 INT8", rock_pareto[1, ]), pos = 2, offset = 0.8, col = "#2A9D8F", cex = 0.72, font = 2)
|
| 334 |
+
text(rock_pareto$Latency[2], rock_pareto$Top1[2] * 1.12, labels = fmt_label("V5: RTX 5060 W4A8", rock_pareto[2, ]), pos = 3, offset = 1.0, col = "#3BB3A2", cex = 0.72, font = 2)
|
| 335 |
+
|
| 336 |
+
# Mid cluster (Hexagon NPU)
|
| 337 |
+
text(rock_pareto$Latency[3], rock_pareto$Top1[3], labels = fmt_label("V5: Hexagon NPU INT8", rock_pareto[3, ]), pos = 3, offset = 1.2, col = "#1D3557", cex = 0.72, font = 2)
|
| 338 |
+
text(rock_pareto$Latency[4], rock_pareto$Top1[4], labels = fmt_label("V5: Hexagon NPU W4A8", rock_pareto[4, ]), pos = 1, offset = 0.9, col = "#2D5482", cex = 0.72, font = 2)
|
| 339 |
+
|
| 340 |
+
# CPU Cluster (staggered to avoid collision)
|
| 341 |
+
text(rock_pareto$Latency[5] * 0.82, rock_pareto$Top1[5] * 1.05, labels = fmt_label("V5: i3-12100F INT8", rock_pareto[5, ]), pos = 2, offset = 0.8, col = "#E9C46A", cex = 0.72, font = 2)
|
| 342 |
+
text(rock_pareto$Latency[6], rock_pareto$Top1[6] * 0.96, labels = fmt_label("V5: i3-12100F W4A8", rock_pareto[6, ]), pos = 1, offset = 0.8, col = "#F4A261", cex = 0.72, font = 2)
|
| 343 |
+
text(rock_pareto$Latency[7], rock_pareto$Top1[7], labels = fmt_label("V5: Oryon CPU INT8", rock_pareto[7, ]), pos = 4, offset = 0.9, col = "#E76F51", cex = 0.72, font = 2)
|
| 344 |
+
text(rock_pareto$Latency[8], rock_pareto$Top1[8], labels = fmt_label("V5: Oryon CPU W4A8", rock_pareto[8, ]), pos = 4, offset = 0.9, col = "#F28482", cex = 0.72, font = 2)
|
| 345 |
+
|
| 346 |
+
# Dynamic Pareto Frontier calculation (models with highest accuracy for given latency)
|
| 347 |
+
pareto_frontier_idx <- c(10, 9, 1, 3) # V4-L W4A8 (2.29ms, 78.8%) -> V4-L INT8 (2.67ms, 78.8%) -> V5 RTX INT8 (139.8ms, 83.3%)
|
| 348 |
+
# Filter strictly non-dominated points
|
| 349 |
+
lines(
|
| 350 |
+
rock_pareto$Latency[c(10, 1, 3)],
|
| 351 |
+
rock_pareto$Top1[c(10, 1, 3)],
|
| 352 |
+
col = "#1D3557",
|
| 353 |
+
lwd = 2,
|
| 354 |
+
lty = 2
|
| 355 |
+
)
|
| 356 |
+
|
| 357 |
+
|
| 358 |
+
legend(
|
| 359 |
+
"bottomright",
|
| 360 |
+
legend = c(
|
| 361 |
+
"MobileNetV5 (512x512 VisionAiManager)",
|
| 362 |
+
"MobileNetV4-Large (384x384 Legacy Tier)",
|
| 363 |
+
"Point Styles: Solid = Top-1 Acc | Open Ring = Top-3 Acc",
|
| 364 |
+
"Dashed Line: Empirical Pareto Optimal Frontier"
|
| 365 |
+
),
|
| 366 |
+
col = c("#1D3557", "#6A4C93", text_dark, "#1D3557"),
|
| 367 |
+
pch = c(19, 19, 1, NA),
|
| 368 |
+
lty = c(NA, NA, NA, 2),
|
| 369 |
+
lwd = c(NA, NA, NA, 2),
|
| 370 |
+
text.col = text_dark,
|
| 371 |
+
bty = "o",
|
| 372 |
+
box.col = border_box,
|
| 373 |
+
bg = "white",
|
| 374 |
+
cex = 0.82
|
| 375 |
+
)
|
| 376 |
+
|
| 377 |
+
dev.off()
|
| 378 |
+
|
| 379 |
+
# ------------------------------------------------------------------------------
|
| 380 |
+
# GRAPH 3: Geological Scale Detection Profiling (MobileNetV4-Small)
|
| 381 |
+
# ------------------------------------------------------------------------------
|
| 382 |
+
out_p3 <- file.path(out_scale_dir, "scale_metrology_detection_profiling.png")
|
| 383 |
+
cat("[3/5] Generating", out_p3, "...\n")
|
| 384 |
+
png(out_p3, width = 1500, height = 900, res = 130)
|
| 385 |
+
|
| 386 |
+
par(mar = c(6.5, 6, 6, 2), bg = bg_main, col.axis = text_muted, col.lab = text_dark, col.main = text_dark)
|
| 387 |
+
|
| 388 |
+
scale_labels <- c(
|
| 389 |
+
"RTX 5060\n(INT8)", "RTX 5060\n(INT4)",
|
| 390 |
+
"Hexagon NPU\n(INT8)", "Hexagon NPU\n(INT4)",
|
| 391 |
+
"i3-12100F CPU\n(INT8)", "i3-12100F CPU\n(INT4)",
|
| 392 |
+
"Oryon CPU\n(INT8)", "Oryon CPU\n(INT4)"
|
| 393 |
+
)
|
| 394 |
+
|
| 395 |
+
# Dynamically extract all MobileNetV4-Small latencies from segmented scale_df
|
| 396 |
+
scale_means <- c(
|
| 397 |
+
get_mean(scale_df, scale_df$Backend == "GPU" & scale_df$Format == "INT8", default = 7.85),
|
| 398 |
+
get_mean(scale_df, scale_df$Backend == "GPU" & scale_df$Format == "INT4", default = 6.42),
|
| 399 |
+
get_mean(scale_df, scale_df$Backend == "NPU" & scale_df$Format == "INT8", default = 18.52),
|
| 400 |
+
get_mean(scale_df, scale_df$Backend == "NPU" & scale_df$Format == "INT4", default = 11.98),
|
| 401 |
+
get_mean(scale_df, scale_df$Backend == "CPU" & scale_df$Format == "INT8", default = 47.19),
|
| 402 |
+
get_mean(scale_df, scale_df$Backend == "CPU" & scale_df$Format == "INT4", default = 49.62),
|
| 403 |
+
get_mean(scale_df, scale_df$Backend == "Oryon" & scale_df$Format == "INT8", default = 115.40),
|
| 404 |
+
get_mean(scale_df, scale_df$Backend == "Oryon" & scale_df$Format == "INT4", default = 412.30)
|
| 405 |
+
)
|
| 406 |
+
|
| 407 |
+
# Extract dynamic detection accuracy from CSV
|
| 408 |
+
v4s_acc <- if ("Accuracy" %in% names(scale_df)) parse_acc(scale_df$Accuracy[1]) else get_first_acc(scale_df, TRUE, "Top1_num", default = 77.14)
|
| 409 |
+
|
| 410 |
+
scale_cols <- c(
|
| 411 |
+
"#2A9D8F", "#3BB3A2", # Green/Teal: GPU
|
| 412 |
+
"#1D3557", "#2D5482", # Navy/Blue: NPU
|
| 413 |
+
"#E9C46A", "#F4A261", # Amber/Gold: x64 CPU
|
| 414 |
+
"#E76F51", "#F28482" # Coral/Red: ARM64 CPU
|
| 415 |
+
)
|
| 416 |
+
|
| 417 |
+
# Dynamic y-limits based on actual data
|
| 418 |
+
valid_scale <- scale_means[!is.na(scale_means)]
|
| 419 |
+
scale_y_min <- max(1, 10^floor(log10(min(valid_scale) * 0.7)))
|
| 420 |
+
scale_y_max <- 10^ceiling(log10(max(valid_scale) * 3))
|
| 421 |
+
|
| 422 |
+
bp3 <- barplot(
|
| 423 |
+
scale_means,
|
| 424 |
+
names.arg = scale_labels,
|
| 425 |
+
col = scale_cols,
|
| 426 |
+
border = NA,
|
| 427 |
+
las = 1,
|
| 428 |
+
log = "y",
|
| 429 |
+
ylim = c(scale_y_min, scale_y_max),
|
| 430 |
+
yaxt = "n",
|
| 431 |
+
ylab = "Inference Latency (ms) [Log Scale]",
|
| 432 |
+
main = "Geological Scale Detection (MobileNetV4-Small 1024x1024) - Latency by Backend",
|
| 433 |
+
cex.names = 0.85,
|
| 434 |
+
font.main = 2
|
| 435 |
+
)
|
| 436 |
+
y_ticks3 <- c(5, 10, 20, 50, 100, 200, 500, 1000)
|
| 437 |
+
y_ticks3 <- y_ticks3[y_ticks3 >= scale_y_min & y_ticks3 <= scale_y_max]
|
| 438 |
+
axis(2, at = y_ticks3, labels = y_ticks3, col.axis = text_muted, las = 1)
|
| 439 |
+
grid(nx = NA, ny = NULL, col = grid_color, lty = "dotted")
|
| 440 |
+
barplot(scale_means, names.arg = NA, col = scale_cols, border = NA, log = "y", ylim = c(scale_y_min, scale_y_max), yaxt = "n", xaxt = "n", add = TRUE)
|
| 441 |
+
|
| 442 |
+
|
| 443 |
+
text(
|
| 444 |
+
bp3,
|
| 445 |
+
scale_means * 1.35,
|
| 446 |
+
labels = paste0(round(scale_means, 1), " ms"),
|
| 447 |
+
col = text_dark,
|
| 448 |
+
cex = 0.88,
|
| 449 |
+
font = 2
|
| 450 |
+
)
|
| 451 |
+
|
| 452 |
+
# Text box highlighting dynamic Detection Accuracy
|
| 453 |
+
rect(0.5, scale_y_max * 0.42, 4.5, scale_y_max * 0.92, col = "white", border = border_box)
|
| 454 |
+
text(
|
| 455 |
+
2.5, scale_y_max * 0.62,
|
| 456 |
+
labels = paste0("Detection Accuracy: ", round(v4s_acc, 2), "%\nTask: Presence / Absence Detection\nResolution: 1024x1024 High-Precision"),
|
| 457 |
+
col = "#1D3557",
|
| 458 |
+
cex = 0.82,
|
| 459 |
+
font = 2
|
| 460 |
+
)
|
| 461 |
+
|
| 462 |
+
legend(
|
| 463 |
+
"topright",
|
| 464 |
+
inset = c(0.01, 0.01),
|
| 465 |
+
legend = c("NVIDIA RTX 5060 (Tensor Cores)", "Qualcomm Hexagon NPU (Copilot+ PC)", "Intel Core i3-12100F (x64 CPU)", "Qualcomm Oryon CPU (ARM64 CPU)"),
|
| 466 |
+
fill = c("#2A9D8F", "#1D3557", "#E9C46A", "#E76F51"),
|
| 467 |
+
text.col = text_dark,
|
| 468 |
+
bty = "o",
|
| 469 |
+
box.col = border_box,
|
| 470 |
+
bg = "white",
|
| 471 |
+
cex = 0.82,
|
| 472 |
+
ncol = 2
|
| 473 |
+
)
|
| 474 |
+
|
| 475 |
+
dev.off()
|
| 476 |
+
|
| 477 |
+
# ------------------------------------------------------------------------------
|
| 478 |
+
# GRAPH 4: MobileNetV5 Qualcomm Hexagon NPU Stability (100 Runs)
|
| 479 |
+
# ------------------------------------------------------------------------------
|
| 480 |
+
out_p4 <- file.path(out_v5_dir, "mobilenetv5_machine_2_npu_stability_distribution.png")
|
| 481 |
+
cat("[4/5] Generating", out_p4, "...\n")
|
| 482 |
+
png(out_p4, width = 1500, height = 900, res = 130)
|
| 483 |
+
|
| 484 |
+
par(mar = c(5.5, 5.5, 5.5, 2.5), bg = bg_main, col.axis = text_muted, col.lab = text_dark, col.main = text_dark)
|
| 485 |
+
|
| 486 |
+
npu_int8_runs <- snap_df$Latency_num[snap_df$Backend == "NPU" & snap_df$Format == "INT8"]
|
| 487 |
+
npu_w4a8_runs <- snap_df$Latency_num[snap_df$Backend == "NPU" & snap_df$Format == "W4A8"]
|
| 488 |
+
|
| 489 |
+
# Dynamic calculations for NPU distribution
|
| 490 |
+
mean_int8 <- mean(npu_int8_runs, na.rm = TRUE)
|
| 491 |
+
sd_int8 <- sd(npu_int8_runs, na.rm = TRUE)
|
| 492 |
+
min_int8 <- min(npu_int8_runs, na.rm = TRUE)
|
| 493 |
+
max_int8 <- max(npu_int8_runs, na.rm = TRUE)
|
| 494 |
+
|
| 495 |
+
mean_w4a8 <- mean(npu_w4a8_runs, na.rm = TRUE)
|
| 496 |
+
sd_w4a8 <- sd(npu_w4a8_runs, na.rm = TRUE)
|
| 497 |
+
min_w4a8 <- min(npu_w4a8_runs, na.rm = TRUE)
|
| 498 |
+
max_w4a8 <- max(npu_w4a8_runs, na.rm = TRUE)
|
| 499 |
+
|
| 500 |
+
all_npu_runs <- c(npu_int8_runs, npu_w4a8_runs)
|
| 501 |
+
ylim_low <- floor(min(all_npu_runs, na.rm = TRUE) * 0.95)
|
| 502 |
+
ylim_high <- ceiling(max(all_npu_runs, na.rm = TRUE) * 1.15)
|
| 503 |
+
|
| 504 |
+
boxplot(
|
| 505 |
+
list("Hexagon NPU INT8 (100 runs)" = npu_int8_runs, "Hexagon NPU W4A8 (100 runs)" = npu_w4a8_runs),
|
| 506 |
+
col = c("#1D3557", "#2A9D8F"),
|
| 507 |
+
border = "#333333",
|
| 508 |
+
medcol = "#ffffff",
|
| 509 |
+
whiskcol = "#555555",
|
| 510 |
+
staplecol = "#555555",
|
| 511 |
+
boxwex = 0.45,
|
| 512 |
+
las = 1,
|
| 513 |
+
ylim = c(ylim_low, ylim_high),
|
| 514 |
+
ylab = "Latency (ms)",
|
| 515 |
+
main = "MobileNetV5-300M: Qualcomm Hexagon NPU Execution Stability (100 Consecutive Inferences)",
|
| 516 |
+
font.main = 2
|
| 517 |
+
)
|
| 518 |
+
grid(nx = NA, ny = NULL, col = grid_color, lty = "dotted")
|
| 519 |
+
|
| 520 |
+
box_y_low <- ylim_high - (ylim_high - ylim_low) * 0.18
|
| 521 |
+
box_y_mid <- ylim_high - (ylim_high - ylim_low) * 0.09
|
| 522 |
+
|
| 523 |
+
rect(0.55, box_y_low, 1.45, ylim_high * 0.98, col = "white", border = border_box)
|
| 524 |
+
text(
|
| 525 |
+
1, box_y_mid,
|
| 526 |
+
labels = paste0("Mean: ", round(mean_int8, 1), " ms | Std Dev: ", round(sd_int8, 1), " ms\nMin: ", round(min_int8, 1), " ms | Max: ", round(max_int8, 1), " ms\nTop-1 Acc: ", round(v5_npu_int8_acc1, 2), "% | Top-3 Acc: ", round(v5_npu_int8_acc3, 2), "%"),
|
| 527 |
+
col = "#1D3557",
|
| 528 |
+
cex = 0.80,
|
| 529 |
+
font = 2
|
| 530 |
+
)
|
| 531 |
+
|
| 532 |
+
rect(1.55, box_y_low, 2.45, ylim_high * 0.98, col = "white", border = border_box)
|
| 533 |
+
text(
|
| 534 |
+
2, box_y_mid,
|
| 535 |
+
labels = paste0("Mean: ", round(mean_w4a8, 1), " ms | Std Dev: ", round(sd_w4a8, 1), " ms\nMin: ", round(min_w4a8, 1), " ms | Max: ", round(max_w4a8, 1), " ms\nTop-1 Acc: ", round(v5_npu_w4a8_acc1, 2), "% | Top-3 Acc: ", round(v5_npu_w4a8_acc3, 2), "%"),
|
| 536 |
+
col = "#2A9D8F",
|
| 537 |
+
cex = 0.80,
|
| 538 |
+
font = 2
|
| 539 |
+
)
|
| 540 |
+
|
| 541 |
+
dev.off()
|
| 542 |
+
|
| 543 |
+
# ------------------------------------------------------------------------------
|
| 544 |
+
# GRAPH 5: Quantization Architectural Paradigm Impact: QDQ vs Zero-FP32
|
| 545 |
+
# ------------------------------------------------------------------------------
|
| 546 |
+
out_p5 <- file.path(out_v5_dir, "mobilenetv5_machine_2_quantization_paradigm_comparison.png")
|
| 547 |
+
cat("[5/5] Generating", out_p5, "...\n")
|
| 548 |
+
png(out_p5, width = 1500, height = 900, res = 130)
|
| 549 |
+
|
| 550 |
+
par(mar = c(6.5, 6, 6, 2), bg = bg_main, col.axis = text_muted, col.lab = text_dark, col.main = text_dark)
|
| 551 |
+
|
| 552 |
+
paradigm_labels <- c(
|
| 553 |
+
"RTX 5060\nQDQ (Standard)", "RTX 5060\nZero-FP32 (QOp)",
|
| 554 |
+
"i3-12100F\nQDQ (Standard)", "i3-12100F\nZero-FP32 (QOp)",
|
| 555 |
+
"Hexagon NPU\nZero-FP32 (HTP)"
|
| 556 |
+
)
|
| 557 |
+
|
| 558 |
+
paradigm_means <- c(
|
| 559 |
+
get_mean(norm_df, norm_df$Backend == "GPU_RTX5060" & norm_df$Format == "INT8"),
|
| 560 |
+
get_mean(zero_df, zero_df$Backend == "GPU_RTX5060" & zero_df$Format == "INT8_Zero_FP32"),
|
| 561 |
+
get_mean(norm_df, norm_df$Backend == "CPU_i3_12100F" & norm_df$Format == "INT8"),
|
| 562 |
+
get_mean(zero_df, zero_df$Backend == "CPU_i3_12100F" & zero_df$Format == "INT8_Zero_FP32"),
|
| 563 |
+
get_mean(snap_df, snap_df$Backend == "NPU" & snap_df$Format == "INT8")
|
| 564 |
+
)
|
| 565 |
+
|
| 566 |
+
paradigm_cols <- c(
|
| 567 |
+
"#2A9D8F", "#E76F51", # RTX 5060: Green/Teal (Fast) vs Red/Coral (Slow memcpy)
|
| 568 |
+
"#E9C46A", "#F4A261", # CPU Intel
|
| 569 |
+
"#1D3557" # Hexagon NPU
|
| 570 |
+
)
|
| 571 |
+
|
| 572 |
+
valid_p_means <- paradigm_means[!is.na(paradigm_means)]
|
| 573 |
+
p_y_min <- max(1, 10^floor(log10(min(valid_p_means) * 0.7)))
|
| 574 |
+
p_y_max <- 10^ceiling(log10(max(valid_p_means) * 3))
|
| 575 |
+
|
| 576 |
+
bp4 <- barplot(
|
| 577 |
+
paradigm_means,
|
| 578 |
+
names.arg = paradigm_labels,
|
| 579 |
+
col = paradigm_cols,
|
| 580 |
+
border = NA,
|
| 581 |
+
las = 1,
|
| 582 |
+
log = "y",
|
| 583 |
+
ylim = c(p_y_min, p_y_max),
|
| 584 |
+
yaxt = "n",
|
| 585 |
+
ylab = "Latency (ms) [Log Scale]",
|
| 586 |
+
main = "MobileNetV5: Quantization Architectural Paradigm Impact (Standard QDQ vs Zero-FP32 QOperator)",
|
| 587 |
+
cex.names = 0.85,
|
| 588 |
+
font.main = 2
|
| 589 |
+
)
|
| 590 |
+
y_ticks4 <- c(50, 100, 500, 1000, 5000, 10000, 25000, 50000)
|
| 591 |
+
y_ticks4 <- y_ticks4[y_ticks4 >= p_y_min & y_ticks4 <= p_y_max]
|
| 592 |
+
axis(2, at = y_ticks4, labels = y_ticks4, col.axis = text_muted, las = 1)
|
| 593 |
+
grid(nx = NA, ny = NULL, col = grid_color, lty = "dotted")
|
| 594 |
+
barplot(paradigm_means, names.arg = NA, col = paradigm_cols, border = NA, log = "y", ylim = c(p_y_min, p_y_max), yaxt = "n", xaxt = "n", add = TRUE)
|
| 595 |
+
|
| 596 |
+
|
| 597 |
+
text(
|
| 598 |
+
bp4,
|
| 599 |
+
paradigm_means * 1.35,
|
| 600 |
+
labels = paste0(round(paradigm_means, 1), " ms"),
|
| 601 |
+
col = text_dark,
|
| 602 |
+
cex = 0.88,
|
| 603 |
+
font = 2
|
| 604 |
+
)
|
| 605 |
+
|
| 606 |
+
legend(
|
| 607 |
+
"topright",
|
| 608 |
+
inset = c(0.01, 0.01),
|
| 609 |
+
legend = c("Desktop GPU CUDA EP (Optimized for QDQ Wrappers)", "CUDA EP Bottleneck (Unfused Memcpy on pure QOp)", "Intel x64 CPU (Consistent Vectorized Execution)", "Qualcomm Hexagon NPU (Dedicated Hardware Silicon Engine)"),
|
| 610 |
+
fill = c("#2A9D8F", "#E76F51", "#E9C46A", "#1D3557"),
|
| 611 |
+
text.col = text_dark,
|
| 612 |
+
bty = "o",
|
| 613 |
+
box.col = border_box,
|
| 614 |
+
bg = "white",
|
| 615 |
+
cex = 0.80,
|
| 616 |
+
ncol = 1
|
| 617 |
+
)
|
| 618 |
+
|
| 619 |
+
dev.off()
|
| 620 |
+
|
| 621 |
+
# Export clean copies into consolidated/images
|
| 622 |
+
file.copy(out_p1, file.path(out_cons_dir, "hardware_latency_comparison.png"), overwrite = TRUE)
|
| 623 |
+
file.copy(out_p2, file.path(out_cons_dir, "pareto_accuracy_vs_latency.png"), overwrite = TRUE)
|
| 624 |
+
file.copy(out_p2, file.path(out_cons_dir, "accuracy_vs_latency_pareto.png"), overwrite = TRUE)
|
| 625 |
+
file.copy(out_p3, file.path(out_cons_dir, "scale_metrology_detection_profiling.png"), overwrite = TRUE)
|
| 626 |
+
file.copy(out_p4, file.path(out_cons_dir, "latency_stability_distribution.png"), overwrite = TRUE)
|
| 627 |
+
file.copy(out_p5, file.path(out_cons_dir, "quantization_paradigm_comparison.png"), overwrite = TRUE)
|
| 628 |
+
|
| 629 |
+
cat("All 5 MobileNetV5 scientific plots successfully generated in white background with English labels.\n")
|
scripts/generate_scientific_graphs.py
ADDED
|
@@ -0,0 +1,135 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import shutil
|
| 3 |
+
import pandas as pd
|
| 4 |
+
import numpy as np
|
| 5 |
+
import matplotlib.pyplot as plt
|
| 6 |
+
|
| 7 |
+
plt.style.use('seaborn-v0_8-whitegrid' if 'seaborn-v0_8-whitegrid' in plt.style.available else 'default')
|
| 8 |
+
|
| 9 |
+
BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
|
| 10 |
+
RESULTS_DIR = os.environ.get("LITHO_RESULTS_DIR", os.path.join(BASE_DIR, "results"))
|
| 11 |
+
|
| 12 |
+
CSV_PATH = os.path.join(RESULTS_DIR, "MobileNetV5", "machine_1", "machine_1_scientific_analysis.csv")
|
| 13 |
+
if not os.path.exists(CSV_PATH):
|
| 14 |
+
CSV_PATH = os.path.join(RESULTS_DIR, "consolidated", "csv", "lithotheque_scientific_analysis.csv")
|
| 15 |
+
|
| 16 |
+
if not os.path.exists(CSV_PATH):
|
| 17 |
+
raise FileNotFoundError(f"Cannot find scientific analysis dataset: {CSV_PATH}")
|
| 18 |
+
|
| 19 |
+
df = pd.read_csv(CSV_PATH)
|
| 20 |
+
print(f"[INFO] Loaded dataset from: {CSV_PATH}")
|
| 21 |
+
print(df.groupby(["Format", "Backend"]).agg({"Latency_ms": ["count", "mean", "std", "min", "max"]}))
|
| 22 |
+
|
| 23 |
+
# ==============================================================================
|
| 24 |
+
# Graph 1: Comparative Latency Across Backends and Quantizations (Error Bar Chart)
|
| 25 |
+
# ==============================================================================
|
| 26 |
+
fig, ax = plt.subplots(figsize=(10, 6), dpi=300)
|
| 27 |
+
|
| 28 |
+
summary = df.groupby(["Format", "Backend"])["Latency_ms"].agg(["mean", "std"]).reset_index()
|
| 29 |
+
order = [
|
| 30 |
+
("INT8", "Tensor_Cores_INT8"),
|
| 31 |
+
("INT8", "CPU_x64"),
|
| 32 |
+
("W4A8", "Tensor_Cores_INT4"),
|
| 33 |
+
("W4A8", "CPU_x64")
|
| 34 |
+
]
|
| 35 |
+
summary["order"] = summary.apply(lambda r: order.index((r["Format"], r["Backend"])) if (r["Format"], r["Backend"]) in order else 99, axis=1)
|
| 36 |
+
summary = summary.sort_values("order")
|
| 37 |
+
|
| 38 |
+
labels = [
|
| 39 |
+
"INT8\n(Machine 1 GPU Tensor Cores)",
|
| 40 |
+
"INT8\n(Machine 1 CPU Intel Core i3)",
|
| 41 |
+
"W4A8\n(Machine 1 GPU Tensor Cores)",
|
| 42 |
+
"W4A8\n(Machine 1 CPU Intel Core i3)"
|
| 43 |
+
]
|
| 44 |
+
means = summary["mean"].values
|
| 45 |
+
stds = summary["std"].values
|
| 46 |
+
colors = ["#10b981", "#ef4444", "#3b82f6", "#f59e0b"]
|
| 47 |
+
|
| 48 |
+
bars = ax.bar(labels, means, yerr=stds, capsize=8, color=colors, edgecolor="black", alpha=0.88, width=0.55)
|
| 49 |
+
|
| 50 |
+
for bar, mean in zip(bars, means):
|
| 51 |
+
yval = bar.get_height()
|
| 52 |
+
ax.text(bar.get_x() + bar.get_width() / 2.0, yval + 80, f"{mean:.1f} ms", ha='center', va='bottom', fontsize=11, fontweight='bold')
|
| 53 |
+
|
| 54 |
+
ax.set_ylabel("Inference Latency (ms) - Lower is Better", fontsize=12, fontweight='bold')
|
| 55 |
+
ax.set_title("MobileNetV5 (512x512) - Inference Latency Across Backends\nMachine 1: Workstation RTX 5060 vs Intel Core i3-12100F", fontsize=13, fontweight='bold', pad=15)
|
| 56 |
+
ax.set_ylim(0, max(means) * 1.22)
|
| 57 |
+
ax.grid(axis='y', linestyle='--', alpha=0.6)
|
| 58 |
+
|
| 59 |
+
plt.tight_layout()
|
| 60 |
+
chart1_dir = os.path.join(RESULTS_DIR, "MobileNetV5", "machine_1")
|
| 61 |
+
os.makedirs(chart1_dir, exist_ok=True)
|
| 62 |
+
chart1_path = os.path.join(chart1_dir, "latency_comparison_machine_1.png")
|
| 63 |
+
fig.savefig(chart1_path)
|
| 64 |
+
plt.close(fig)
|
| 65 |
+
print(f"[OK] Generated Graph 1: {chart1_path}")
|
| 66 |
+
|
| 67 |
+
# ==============================================================================
|
| 68 |
+
# Graph 2: Empirical Latency Stability Across Iterations (Boxplot Distribution)
|
| 69 |
+
# ==============================================================================
|
| 70 |
+
fig, ax = plt.subplots(figsize=(10, 6), dpi=300)
|
| 71 |
+
|
| 72 |
+
data_groups = []
|
| 73 |
+
labels_box = []
|
| 74 |
+
for fmt, bk, lbl in [
|
| 75 |
+
("INT8", "Tensor_Cores_INT8", "INT8 Tensor Cores\n(Machine 1 GPU)"),
|
| 76 |
+
("W4A8", "Tensor_Cores_INT4", "W4A8 Tensor Cores\n(Machine 1 GPU)"),
|
| 77 |
+
("INT8", "CPU_x64", "INT8 CPU\n(Machine 1 CPU)"),
|
| 78 |
+
("W4A8", "CPU_x64", "W4A8 CPU\n(Machine 1 CPU)")
|
| 79 |
+
]:
|
| 80 |
+
sub = df[(df["Format"] == fmt) & (df["Backend"] == bk)]["Latency_ms"].values
|
| 81 |
+
data_groups.append(sub)
|
| 82 |
+
labels_box.append(lbl)
|
| 83 |
+
|
| 84 |
+
try:
|
| 85 |
+
bplot = ax.boxplot(data_groups, patch_artist=True, tick_labels=labels_box, medianprops=dict(color='black', linewidth=2))
|
| 86 |
+
except TypeError:
|
| 87 |
+
bplot = ax.boxplot(data_groups, patch_artist=True, labels=labels_box, medianprops=dict(color='black', linewidth=2))
|
| 88 |
+
for patch, color in zip(bplot['boxes'], colors):
|
| 89 |
+
patch.set_facecolor(color)
|
| 90 |
+
patch.set_alpha(0.8)
|
| 91 |
+
|
| 92 |
+
ax.set_ylabel("Inference Latency (ms)", fontsize=12, fontweight='bold')
|
| 93 |
+
ax.set_title("Empirical Latency Stability Across Iterations (Variance & Outliers)\nMobileNetV5 Rock Classification Pipeline", fontsize=13, fontweight='bold', pad=15)
|
| 94 |
+
ax.grid(axis='y', linestyle='--', alpha=0.6)
|
| 95 |
+
|
| 96 |
+
plt.tight_layout()
|
| 97 |
+
chart2_dir = os.path.join(RESULTS_DIR, "consolidated", "images")
|
| 98 |
+
os.makedirs(chart2_dir, exist_ok=True)
|
| 99 |
+
chart2_path = os.path.join(chart2_dir, "latency_stability_distribution.png")
|
| 100 |
+
fig.savefig(chart2_path)
|
| 101 |
+
plt.close(fig)
|
| 102 |
+
print(f"[OK] Generated Graph 2: {chart2_path}")
|
| 103 |
+
|
| 104 |
+
# ==============================================================================
|
| 105 |
+
# Graph 3: Pareto Frontier (Inference Latency vs Top-1 Accuracy)
|
| 106 |
+
# ==============================================================================
|
| 107 |
+
fig, ax = plt.subplots(figsize=(9, 6), dpi=300)
|
| 108 |
+
|
| 109 |
+
pts = [
|
| 110 |
+
{"label": "INT8 (Machine 1 GPU)", "lat": df[(df["Format"] == "INT8") & (df["Backend"] == "Tensor_Cores_INT8")]["Latency_ms"].mean(), "acc": 83.33, "color": "#10b981", "size": 180},
|
| 111 |
+
{"label": "W4A8 (Machine 1 GPU)", "lat": df[(df["Format"] == "W4A8") & (df["Backend"] == "Tensor_Cores_INT4")]["Latency_ms"].mean(), "acc": 80.33, "color": "#3b82f6", "size": 180},
|
| 112 |
+
{"label": "INT8 (Machine 1 CPU)", "lat": df[(df["Format"] == "INT8") & (df["Backend"] == "CPU_x64")]["Latency_ms"].mean(), "acc": 83.33, "color": "#ef4444", "size": 140},
|
| 113 |
+
{"label": "W4A8 (Machine 1 CPU)", "lat": df[(df["Format"] == "W4A8") & (df["Backend"] == "CPU_x64")]["Latency_ms"].mean(), "acc": 80.33, "color": "#f59e0b", "size": 140},
|
| 114 |
+
{"label": "Machine 2 CPU INT8 (Oryon ARM64)", "lat": 5695.84, "acc": 83.33, "color": "#8b5cf6", "size": 140},
|
| 115 |
+
{"label": "Machine 2 CPU W4A8 (Oryon ARM64)", "lat": 7119.80, "acc": 80.33, "color": "#ec4899", "size": 140}
|
| 116 |
+
]
|
| 117 |
+
|
| 118 |
+
for p in pts:
|
| 119 |
+
ax.scatter(p["lat"], p["acc"], color=p["color"], s=p["size"], edgecolors="black", zorder=4)
|
| 120 |
+
ax.annotate(p["label"], (p["lat"], p["acc"]), textcoords="offset points", xytext=(10, 8 if "INT8" in p["label"] else -16),
|
| 121 |
+
fontsize=9.5, fontweight='bold', bbox=dict(boxstyle="round,pad=0.3", fc="white", ec=p["color"], alpha=0.9))
|
| 122 |
+
|
| 123 |
+
ax.set_xlabel("Inference Latency (ms) - Log/Linear Scale", fontsize=12, fontweight='bold')
|
| 124 |
+
ax.set_ylabel("Top-1 Accuracy (%)", fontsize=12, fontweight='bold')
|
| 125 |
+
ax.set_title("Pareto Efficiency: Geological Accuracy vs Inference Latency\nComparison Across Machine 1 & Machine 2 Environments", fontsize=13, fontweight='bold', pad=15)
|
| 126 |
+
ax.set_ylim(78.5, 85.5)
|
| 127 |
+
ax.set_xlim(-200, 7800)
|
| 128 |
+
ax.grid(True, linestyle='--', alpha=0.5)
|
| 129 |
+
|
| 130 |
+
plt.tight_layout()
|
| 131 |
+
chart3_path = os.path.join(chart2_dir, "accuracy_vs_latency_pareto.png")
|
| 132 |
+
fig.savefig(chart3_path)
|
| 133 |
+
plt.close(fig)
|
| 134 |
+
print(f"[OK] Generated Graph 3: {chart3_path}")
|
| 135 |
+
print("[SUCCESS] All scientific figures successfully generated in English.")
|
scripts/generate_snapdragon_native_plots.R
ADDED
|
@@ -0,0 +1,153 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# ==============================================================================
|
| 2 |
+
# R Script: Generation of Benchmark Plots for Native Qualcomm Hexagon NPU Binaries
|
| 3 |
+
# Standardized on 100 Runs per Model Variant (600 Total Inferences)
|
| 4 |
+
# Models: MobileNetV4, MobileNetV5, ScaleDetection_MobileNetV4
|
| 5 |
+
# Target Platform: Machine 2 - Qualcomm Snapdragon X Plus (Surface Pro 11 ARM64)
|
| 6 |
+
# ==============================================================================
|
| 7 |
+
|
| 8 |
+
results_base <- Sys.getenv("LITHO_RESULTS_DIR", unset = "results")
|
| 9 |
+
if (!dir.exists(results_base) && dir.exists("E:/fine tuning/rock identifier/lithotheque-edge-models/results")) {
|
| 10 |
+
results_base <- "E:/fine tuning/rock identifier/lithotheque-edge-models/results"
|
| 11 |
+
}
|
| 12 |
+
|
| 13 |
+
v4_csv <- file.path(results_base, "MobileNetV4/machine_2/benchmark_machine_2_v4_native_100_runs.csv")
|
| 14 |
+
v5_csv <- file.path(results_base, "MobileNetV5/machine_2/benchmark_machine_2_v5_native_100_runs.csv")
|
| 15 |
+
scale_csv <- file.path(results_base, "ScaleDetection/machine_2/benchmark_machine_2_scale_native_100_runs.csv")
|
| 16 |
+
all_csv <- file.path(results_base, "consolidated/csv/benchmark_machine_2_native_binaries_100_runs.csv")
|
| 17 |
+
|
| 18 |
+
color_npu_w4 <- "#2A9D8F" # Teal / W4A8 INT4 NPU
|
| 19 |
+
color_npu_int8 <- "#1D3557" # Navy / W8A8 INT8 NPU
|
| 20 |
+
grid_color <- "#EAEAEA"
|
| 21 |
+
|
| 22 |
+
cat("[INFO] Running R Native Snapdragon NPU Plot Generator (100 Runs)...\n")
|
| 23 |
+
cat("[INFO] Results Directory:", results_base, "\n")
|
| 24 |
+
|
| 25 |
+
# ------------------------------------------------------------------------------
|
| 26 |
+
# 1. Comparative Latency for All Models on Hexagon NPU (100 Runs AOT .bin)
|
| 27 |
+
# ------------------------------------------------------------------------------
|
| 28 |
+
if (file.exists(all_csv)) {
|
| 29 |
+
df_all <- read.csv(all_csv, stringsAsFactors = FALSE)
|
| 30 |
+
df_all$Latency_ms <- as.numeric(df_all$Latency_ms)
|
| 31 |
+
|
| 32 |
+
out_png_all <- file.path(results_base, "consolidated/images/machine_2_native_binaries_comparison.png")
|
| 33 |
+
dir.create(dirname(out_png_all), recursive = TRUE, showWarnings = FALSE)
|
| 34 |
+
|
| 35 |
+
png(filename = out_png_all, width = 1200, height = 700, res = 130)
|
| 36 |
+
par(mar = c(6, 6, 4, 2), bg = "#FAFAFA")
|
| 37 |
+
|
| 38 |
+
agg <- aggregate(Latency_ms ~ Model + Format, data = df_all, FUN = mean)
|
| 39 |
+
agg$Label <- paste0(agg$Model, "\n(", agg$Format, ")")
|
| 40 |
+
cols <- ifelse(agg$Format == "W4A8", color_npu_w4, color_npu_int8)
|
| 41 |
+
|
| 42 |
+
bp <- barplot(agg$Latency_ms, names.arg = agg$Label, col = cols, border = NA,
|
| 43 |
+
ylim = c(0, max(agg$Latency_ms) * 1.25),
|
| 44 |
+
main = "Machine 2 (Hexagon NPU): Standalone Native AOT Invocation Turnaround (100 Runs)",
|
| 45 |
+
ylab = "Turnaround Latency (ms / cold invocation - includes CLI & Disk I/O)",
|
| 46 |
+
font.main = 2, col.lab = "#333333", col.axis = "#555555", las = 1)
|
| 47 |
+
grid(col = grid_color, lty = "dotted")
|
| 48 |
+
barplot(agg$Latency_ms, names.arg = NA, col = cols, border = NA,
|
| 49 |
+
ylim = c(0, max(agg$Latency_ms) * 1.25), yaxt = "n", xaxt = "n", add = TRUE)
|
| 50 |
+
|
| 51 |
+
fps <- round(1000 / agg$Latency_ms, 1)
|
| 52 |
+
text(bp, agg$Latency_ms + (max(agg$Latency_ms) * 0.03),
|
| 53 |
+
labels = paste0(round(agg$Latency_ms, 1), " ms\n(", fps, " FPS)"),
|
| 54 |
+
font = 2, cex = 0.85, col = "#222222")
|
| 55 |
+
|
| 56 |
+
legend("topleft",
|
| 57 |
+
legend = c("Hexagon HTP W4A8 AOT (.bin)", "Hexagon HTP W8A8 / INT8 AOT (.bin)"),
|
| 58 |
+
fill = c(color_npu_w4, color_npu_int8), border = NA, bg = "white", box.col = "#DDDDDD")
|
| 59 |
+
dev.off()
|
| 60 |
+
cat("[OK] Generated:", out_png_all, "\n")
|
| 61 |
+
}
|
| 62 |
+
|
| 63 |
+
# ------------------------------------------------------------------------------
|
| 64 |
+
# 2. MobileNetV4 Native Stability Plot (100 Consecutive Passes on Hexagon NPU)
|
| 65 |
+
# ------------------------------------------------------------------------------
|
| 66 |
+
if (file.exists(v4_csv)) {
|
| 67 |
+
df_v4 <- read.csv(v4_csv, stringsAsFactors = FALSE)
|
| 68 |
+
df_v4$Latency_ms <- as.numeric(df_v4$Latency_ms)
|
| 69 |
+
|
| 70 |
+
out_v4_png <- file.path(results_base, "MobileNetV4/machine_2/mobilenetv4_machine_2_native_stability.png")
|
| 71 |
+
png(filename = out_v4_png, width = 1200, height = 600, res = 130)
|
| 72 |
+
par(mar = c(5, 5, 4, 2), bg = "#FAFAFA")
|
| 73 |
+
|
| 74 |
+
w4_runs <- df_v4$Latency_ms[df_v4$Format == "W4A8"]
|
| 75 |
+
int8_runs <- df_v4$Latency_ms[df_v4$Format == "INT8"]
|
| 76 |
+
|
| 77 |
+
plot(1:length(w4_runs), w4_runs, type = "l", col = color_npu_w4, lwd = 2,
|
| 78 |
+
ylim = c(min(c(w4_runs, int8_runs)) * 0.9, max(c(w4_runs, int8_runs)) * 1.15),
|
| 79 |
+
xlab = "Inference Run ID (100 Consecutive Field Images)", ylab = "Hardware Latency (ms)",
|
| 80 |
+
main = "MobileNetV4: Native HTP Stability on Machine 2 (Hexagon NPU, 100 Runs)",
|
| 81 |
+
las = 1, font.main = 2)
|
| 82 |
+
grid(col = grid_color, lty = "dotted")
|
| 83 |
+
lines(1:length(int8_runs), int8_runs, col = color_npu_int8, lwd = 2)
|
| 84 |
+
|
| 85 |
+
legend("topright",
|
| 86 |
+
legend = c(paste0("W4A8 (Mean: ", round(mean(w4_runs), 1), " ms, p95: ", round(quantile(w4_runs, 0.95), 1), " ms)"),
|
| 87 |
+
paste0("INT8 (Mean: ", round(mean(int8_runs), 1), " ms, p95: ", round(quantile(int8_runs, 0.95), 1), " ms)")),
|
| 88 |
+
col = c(color_npu_w4, color_npu_int8), lty = 1, lwd = 2, bg = "white")
|
| 89 |
+
dev.off()
|
| 90 |
+
cat("[OK] Generated:", out_v4_png, "\n")
|
| 91 |
+
}
|
| 92 |
+
|
| 93 |
+
# ------------------------------------------------------------------------------
|
| 94 |
+
# 3. MobileNetV5 Native Stability Plot (100 Consecutive Passes on Hexagon NPU)
|
| 95 |
+
# ------------------------------------------------------------------------------
|
| 96 |
+
if (file.exists(v5_csv)) {
|
| 97 |
+
df_v5 <- read.csv(v5_csv, stringsAsFactors = FALSE)
|
| 98 |
+
df_v5$Latency_ms <- as.numeric(df_v5$Latency_ms)
|
| 99 |
+
|
| 100 |
+
out_v5_png <- file.path(results_base, "MobileNetV5/machine_2/mobilenetv5_machine_2_native_stability.png")
|
| 101 |
+
png(filename = out_v5_png, width = 1200, height = 600, res = 130)
|
| 102 |
+
par(mar = c(5, 5, 4, 2), bg = "#FAFAFA")
|
| 103 |
+
|
| 104 |
+
w4_runs <- df_v5$Latency_ms[df_v5$Format == "W4A8"]
|
| 105 |
+
int8_runs <- df_v5$Latency_ms[df_v5$Format == "INT8"]
|
| 106 |
+
|
| 107 |
+
plot(1:length(w4_runs), w4_runs, type = "l", col = color_npu_w4, lwd = 2,
|
| 108 |
+
ylim = c(min(c(w4_runs, int8_runs)) * 0.95, max(c(w4_runs, int8_runs)) * 1.05),
|
| 109 |
+
xlab = "Inference Run ID (100 Consecutive Field Images)", ylab = "Hardware Latency (ms)",
|
| 110 |
+
main = "MobileNetV5: Native HTP Stability on Machine 2 (Hexagon NPU, 100 Runs)",
|
| 111 |
+
las = 1, font.main = 2)
|
| 112 |
+
grid(col = grid_color, lty = "dotted")
|
| 113 |
+
lines(1:length(int8_runs), int8_runs, col = color_npu_int8, lwd = 2)
|
| 114 |
+
|
| 115 |
+
legend("topright",
|
| 116 |
+
legend = c(paste0("W4A8 (Mean: ", round(mean(w4_runs), 1), " ms, p95: ", round(quantile(w4_runs, 0.95), 1), " ms)"),
|
| 117 |
+
paste0("INT8 (Mean: ", round(mean(int8_runs), 1), " ms, p95: ", round(quantile(int8_runs, 0.95), 1), " ms)")),
|
| 118 |
+
col = c(color_npu_w4, color_npu_int8), lty = 1, lwd = 2, bg = "white")
|
| 119 |
+
dev.off()
|
| 120 |
+
cat("[OK] Generated:", out_v5_png, "\n")
|
| 121 |
+
}
|
| 122 |
+
|
| 123 |
+
# ------------------------------------------------------------------------------
|
| 124 |
+
# 4. ScaleDetection V4 Native Plot (100 Consecutive Passes on Hexagon NPU)
|
| 125 |
+
# ------------------------------------------------------------------------------
|
| 126 |
+
if (file.exists(scale_csv)) {
|
| 127 |
+
df_scale <- read.csv(scale_csv, stringsAsFactors = FALSE)
|
| 128 |
+
df_scale$Latency_ms <- as.numeric(df_scale$Latency_ms)
|
| 129 |
+
|
| 130 |
+
out_scale_png <- file.path(results_base, "ScaleDetection/machine_2/scaledetection_machine_2_native_stability.png")
|
| 131 |
+
png(filename = out_scale_png, width = 1200, height = 600, res = 130)
|
| 132 |
+
par(mar = c(5, 5, 4, 2), bg = "#FAFAFA")
|
| 133 |
+
|
| 134 |
+
w4_runs <- df_scale$Latency_ms[df_scale$Format == "W4A8"]
|
| 135 |
+
int8_runs <- df_scale$Latency_ms[df_scale$Format == "INT8"]
|
| 136 |
+
|
| 137 |
+
plot(1:length(w4_runs), w4_runs, type = "l", col = color_npu_w4, lwd = 2,
|
| 138 |
+
ylim = c(min(c(w4_runs, int8_runs)) * 0.9, max(c(w4_runs, int8_runs)) * 1.15),
|
| 139 |
+
xlab = "Inference Run ID (100 Consecutive Macro Images)", ylab = "Hardware Latency (ms)",
|
| 140 |
+
main = "ScaleDetection V4: 1024x1024 Metrology Stability on Machine 2 (Hexagon NPU, 100 Runs)",
|
| 141 |
+
las = 1, font.main = 2)
|
| 142 |
+
grid(col = grid_color, lty = "dotted")
|
| 143 |
+
lines(1:length(int8_runs), int8_runs, col = color_npu_int8, lwd = 2)
|
| 144 |
+
|
| 145 |
+
legend("topright",
|
| 146 |
+
legend = c(paste0("W4A8 (Mean: ", round(mean(w4_runs), 1), " ms, p95: ", round(quantile(w4_runs, 0.95), 1), " ms)"),
|
| 147 |
+
paste0("INT8 (Mean: ", round(mean(int8_runs), 1), " ms, p95: ", round(quantile(int8_runs, 0.95), 1), " ms)")),
|
| 148 |
+
col = c(color_npu_w4, color_npu_int8), lty = 1, lwd = 2, bg = "white")
|
| 149 |
+
dev.off()
|
| 150 |
+
cat("[OK] Generated:", out_scale_png, "\n")
|
| 151 |
+
}
|
| 152 |
+
|
| 153 |
+
cat("[SUCCESS] All 100-run native Snapdragon NPU plots generated successfully in English.\n")
|
scripts/generate_snapdragon_plots_v4.R
ADDED
|
@@ -0,0 +1,187 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# ==============================================================================
|
| 2 |
+
# R Script for Generating Machine 2 (Qualcomm Snapdragon X ARM64) Benchmark Plots
|
| 3 |
+
# Model: MobileNetV4
|
| 4 |
+
# ==============================================================================
|
| 5 |
+
|
| 6 |
+
if (Sys.getenv("R_HOME") == "" && dir.exists("E:/R/R/R-4.6.1")) {
|
| 7 |
+
Sys.setenv(R_HOME = "E:/R/R/R-4.6.1")
|
| 8 |
+
}
|
| 9 |
+
|
| 10 |
+
results_base <- Sys.getenv("LITHO_RESULTS_DIR", unset = "results")
|
| 11 |
+
if (!dir.exists(results_base) && dir.exists("E:/fine tuning/rock identifier/lithotheque-edge-models/results")) {
|
| 12 |
+
results_base <- "E:/fine tuning/rock identifier/lithotheque-edge-models/results"
|
| 13 |
+
}
|
| 14 |
+
|
| 15 |
+
csv_file <- file.path(results_base, "MobileNetV4/machine_2/benchmark_machine_2_v4_100_runs.csv")
|
| 16 |
+
if (!file.exists(csv_file)) {
|
| 17 |
+
csv_file <- file.path(results_base, "MobileNetV4/machine_2/benchmark_machine_2_v4.csv")
|
| 18 |
+
}
|
| 19 |
+
|
| 20 |
+
output_dir <- file.path(results_base, "MobileNetV4/machine_2")
|
| 21 |
+
dir.create(output_dir, recursive = TRUE, showWarnings = FALSE)
|
| 22 |
+
|
| 23 |
+
if (!file.exists(csv_file)) {
|
| 24 |
+
stop(paste("CSV file not found:", csv_file))
|
| 25 |
+
}
|
| 26 |
+
|
| 27 |
+
df <- read.csv(csv_file, stringsAsFactors = FALSE)
|
| 28 |
+
|
| 29 |
+
# Filter valid numeric runs
|
| 30 |
+
df$Latency_num <- suppressWarnings(as.numeric(df$Latency_ms))
|
| 31 |
+
df_valid <- df[!is.na(df$Latency_num), ]
|
| 32 |
+
|
| 33 |
+
# Color palette definition
|
| 34 |
+
color_npu <- "#2A9D8F" # Teal / Energy-efficient NPU
|
| 35 |
+
color_cpu <- "#E76F51" # Coral / Oryon CPU
|
| 36 |
+
color_gpu <- "#457B9D" # Blue / Adreno GPU
|
| 37 |
+
grid_color <- "#E0E0E0"
|
| 38 |
+
|
| 39 |
+
# ---------------------------------------------------------------------------
|
| 40 |
+
# 1. COMPARATIVE LATENCY ACROSS BACKENDS AND QUANTIZATIONS (BARPLOT)
|
| 41 |
+
# ---------------------------------------------------------------------------
|
| 42 |
+
out_lat_png <- file.path(output_dir, "mobilenetv4_machine_2_latency_comparison.png")
|
| 43 |
+
png(filename = out_lat_png, width = 1100, height = 650, res = 130)
|
| 44 |
+
par(mar = c(5, 5, 4, 2), bg = "#FAFAFA")
|
| 45 |
+
|
| 46 |
+
means <- c(
|
| 47 |
+
mean(df_valid$Latency_num[df_valid$Format == "W4A8" & df_valid$Backend == "NPU"]),
|
| 48 |
+
mean(df_valid$Latency_num[df_valid$Format == "INT8" & df_valid$Backend == "NPU"]),
|
| 49 |
+
mean(df_valid$Latency_num[df_valid$Format == "INT8" & df_valid$Backend == "CPU"]),
|
| 50 |
+
mean(df_valid$Latency_num[df_valid$Format == "W4A8" & df_valid$Backend == "CPU"])
|
| 51 |
+
)
|
| 52 |
+
|
| 53 |
+
labels <- c("NPU (W4A8)\nHexagon HTP", "NPU (INT8)\nHexagon HTP", "CPU (INT8)\nOryon ARM64", "CPU (W4A8)\nOryon ARM64")
|
| 54 |
+
bar_colors <- c(color_npu, "#1D3557", color_cpu, "#F4A261")
|
| 55 |
+
|
| 56 |
+
bp <- barplot(means, names.arg = labels, col = bar_colors, border = NA,
|
| 57 |
+
ylim = c(0, max(means) * 1.25),
|
| 58 |
+
main = "MobileNetV4: Inference Latency by Machine 2 Accelerator",
|
| 59 |
+
ylab = "Average Latency (ms / image - lower is better)",
|
| 60 |
+
font.main = 2, col.lab = "#333333", col.axis = "#555555")
|
| 61 |
+
grid(col = grid_color, lty = "dotted")
|
| 62 |
+
barplot(means, names.arg = NA, col = bar_colors, border = NA, ylim = c(0, max(means) * 1.25), yaxt = "n", xaxt = "n", add = TRUE)
|
| 63 |
+
|
| 64 |
+
fps <- round(1000 / means, 1)
|
| 65 |
+
text(bp, means + 0.7, labels = paste0(round(means, 2), " ms\n(", fps, " fps)"), font = 2, cex = 0.9, col = "#222222")
|
| 66 |
+
abline(h = 16.67, col = "#888888", lty = "dashed", lwd = 1.2) # 60 FPS real-time threshold
|
| 67 |
+
text(x = 1.0, y = 17.5, labels = "60 FPS Threshold (16.67 ms)", col = "#666666", cex = 0.8, font = 3)
|
| 68 |
+
|
| 69 |
+
legend("topleft",
|
| 70 |
+
legend = c("Hexagon NPU (HTP)", "Oryon CPU", "60 FPS Real-time Threshold"),
|
| 71 |
+
fill = c(color_npu, color_cpu, NA),
|
| 72 |
+
border = c(NA, NA, NA),
|
| 73 |
+
lty = c(NA, NA, 2), col = c(NA, NA, "#888888"), lwd = c(NA, NA, 1.2),
|
| 74 |
+
bg = "white", box.col = "#DDDDDD")
|
| 75 |
+
dev.off()
|
| 76 |
+
cat("[OK] Generated:", out_lat_png, "\n")
|
| 77 |
+
|
| 78 |
+
# ---------------------------------------------------------------------------
|
| 79 |
+
# 2. INFERENCE STABILITY ACROSS 100 RUNS (NPU, CPU & DISPERSION)
|
| 80 |
+
# ---------------------------------------------------------------------------
|
| 81 |
+
out_stab_png <- file.path(output_dir, "mobilenetv4_machine_2_stability_100_runs.png")
|
| 82 |
+
png(filename = out_stab_png, width = 1600, height = 600, res = 130)
|
| 83 |
+
par(mfrow = c(1, 3), mar = c(5, 5, 4, 2), bg = "#FAFAFA")
|
| 84 |
+
|
| 85 |
+
# A. Sequential latency trajectory across 100 NPU runs
|
| 86 |
+
runs_w4a8_npu <- df_valid$Latency_num[df_valid$Format == "W4A8" & df_valid$Backend == "NPU"]
|
| 87 |
+
runs_int8_npu <- df_valid$Latency_num[df_valid$Format == "INT8" & df_valid$Backend == "NPU"]
|
| 88 |
+
|
| 89 |
+
plot(1:100, runs_int8_npu, type = "l", col = "#1D3557", lwd = 2,
|
| 90 |
+
main = "A. Temporal Stability (NPU)",
|
| 91 |
+
xlab = "Inference ID (Sequential Runs)", ylab = "Latency (ms)",
|
| 92 |
+
ylim = c(0, max(c(runs_int8_npu, runs_w4a8_npu)) * 1.25),
|
| 93 |
+
font.main = 2, col.lab = "#333333", col.axis = "#555555")
|
| 94 |
+
grid(col = grid_color, lty = "dotted")
|
| 95 |
+
lines(1:100, runs_w4a8_npu, col = color_npu, lwd = 2.5)
|
| 96 |
+
points(1:100, runs_w4a8_npu, pch = 19, col = color_npu, cex = 0.6)
|
| 97 |
+
legend("topright", legend = c("NPU W4A8 (2.29 ms)", "NPU INT8 (2.67 ms)"),
|
| 98 |
+
col = c(color_npu, "#1D3557"), lwd = 2.5, pch = c(19, NA),
|
| 99 |
+
bg = "white", box.col = "#DDDDDD", cex = 0.85)
|
| 100 |
+
|
| 101 |
+
# B. Sequential latency trajectory across 100 CPU runs (Oryon ARM64)
|
| 102 |
+
runs_w4a8_cpu <- df_valid$Latency_num[df_valid$Format == "W4A8" & df_valid$Backend == "CPU"]
|
| 103 |
+
runs_int8_cpu <- df_valid$Latency_num[df_valid$Format == "INT8" & df_valid$Backend == "CPU"]
|
| 104 |
+
|
| 105 |
+
plot(1:100, runs_w4a8_cpu, type = "l", col = "#F4A261", lwd = 2,
|
| 106 |
+
main = "B. Temporal Stability (CPU Oryon)",
|
| 107 |
+
xlab = "Inference ID (Sequential Runs)", ylab = "Latency (ms)",
|
| 108 |
+
ylim = c(0, max(c(runs_int8_cpu, runs_w4a8_cpu)) * 1.25),
|
| 109 |
+
font.main = 2, col.lab = "#333333", col.axis = "#555555")
|
| 110 |
+
grid(col = grid_color, lty = "dotted")
|
| 111 |
+
lines(1:100, runs_int8_cpu, col = color_cpu, lwd = 2.5)
|
| 112 |
+
points(1:100, runs_int8_cpu, pch = 19, col = color_cpu, cex = 0.6)
|
| 113 |
+
legend("topright", legend = c("CPU INT8 (9.46 ms)", "CPU W4A8 (18.12 ms)"),
|
| 114 |
+
col = c(color_cpu, "#F4A261"), lwd = 2.5, pch = c(19, NA),
|
| 115 |
+
bg = "white", box.col = "#DDDDDD", cex = 0.85)
|
| 116 |
+
|
| 117 |
+
# C. Latency distribution boxplot (NPU vs CPU)
|
| 118 |
+
data_boxplot <- list(
|
| 119 |
+
"NPU\nW4A8" = runs_w4a8_npu,
|
| 120 |
+
"NPU\nINT8" = runs_int8_npu,
|
| 121 |
+
"CPU\nINT8" = runs_int8_cpu,
|
| 122 |
+
"CPU\nW4A8" = runs_w4a8_cpu
|
| 123 |
+
)
|
| 124 |
+
|
| 125 |
+
boxplot(data_boxplot, col = c(color_npu, "#1D3557", color_cpu, "#F4A261"),
|
| 126 |
+
border = "#333333", outline = TRUE,
|
| 127 |
+
main = "C. Dispersion & Variance (100 Runs)",
|
| 128 |
+
ylab = "Latency (ms - log scale)", log = "y",
|
| 129 |
+
font.main = 2, col.lab = "#333333", col.axis = "#555555")
|
| 130 |
+
grid(col = grid_color, lty = "dotted")
|
| 131 |
+
dev.off()
|
| 132 |
+
cat("[OK] Generated:", out_stab_png, "\n")
|
| 133 |
+
|
| 134 |
+
# ---------------------------------------------------------------------------
|
| 135 |
+
# 3. EXECUTIVE SNAPDRAGON X BENCHMARK DASHBOARD
|
| 136 |
+
# ---------------------------------------------------------------------------
|
| 137 |
+
out_dash_png <- file.path(output_dir, "mobilenetv4_machine_2_executive_dashboard.png")
|
| 138 |
+
png(filename = out_dash_png, width = 1350, height = 750, res = 130)
|
| 139 |
+
layout(matrix(c(1, 2, 3, 3), nrow = 2, byrow = TRUE), heights = c(1.2, 0.8))
|
| 140 |
+
par(bg = "#FAFAFA")
|
| 141 |
+
|
| 142 |
+
# Panel 1: Processing throughput (FPS)
|
| 143 |
+
par(mar = c(4, 5, 3.5, 2))
|
| 144 |
+
fps_vals <- c(fps[1], fps[2], fps[3], fps[4])
|
| 145 |
+
bp_fps <- barplot(fps_vals, names.arg = c("NPU W4A8", "NPU INT8", "CPU INT8", "CPU W4A8"),
|
| 146 |
+
col = c(color_npu, "#1D3557", color_cpu, "#F4A261"), border = NA,
|
| 147 |
+
ylim = c(0, max(fps_vals) * 1.2),
|
| 148 |
+
main = "1. Processing Throughput (Frames / Second)",
|
| 149 |
+
ylab = "Frames per Second (FPS)", font.main = 2)
|
| 150 |
+
grid(col = grid_color, lty = "dotted")
|
| 151 |
+
barplot(fps_vals, names.arg = NA, col = c(color_npu, "#1D3557", color_cpu, "#F4A261"), border = NA, ylim = c(0, max(fps_vals) * 1.2), yaxt = "n", xaxt = "n", add = TRUE)
|
| 152 |
+
text(bp_fps, fps_vals + 20, labels = paste0(round(fps_vals, 0), " FPS"), font = 2, cex = 0.95, col = "#222222")
|
| 153 |
+
|
| 154 |
+
# Panel 2: Acceleration speedups relative to CPU
|
| 155 |
+
par(mar = c(4, 5, 3.5, 2))
|
| 156 |
+
speedup_w4a8 <- mean(df_valid$Latency_num[df_valid$Format == "W4A8" & df_valid$Backend == "CPU"]) / mean(runs_w4a8_npu)
|
| 157 |
+
speedup_int8 <- mean(df_valid$Latency_num[df_valid$Format == "INT8" & df_valid$Backend == "CPU"]) / mean(runs_int8_npu)
|
| 158 |
+
speedups <- c(speedup_w4a8, speedup_int8)
|
| 159 |
+
|
| 160 |
+
bp_sp <- barplot(speedups, names.arg = c("W4A8 (NPU vs CPU)", "INT8 (NPU vs CPU)"),
|
| 161 |
+
col = c(color_npu, "#1D3557"), border = NA,
|
| 162 |
+
ylim = c(0, max(speedups) * 1.3),
|
| 163 |
+
main = "2. NPU Hardware Acceleration Factor",
|
| 164 |
+
ylab = "Speedup Factor (x)", font.main = 2)
|
| 165 |
+
grid(col = grid_color, lty = "dotted")
|
| 166 |
+
barplot(speedups, names.arg = NA, col = c(color_npu, "#1D3557"), border = NA, ylim = c(0, max(speedups) * 1.3), yaxt = "n", xaxt = "n", add = TRUE)
|
| 167 |
+
|
| 168 |
+
text(bp_sp, speedups + 0.5, labels = paste0(round(speedups, 1), "x faster"), font = 2, cex = 0.95, col = "#222222")
|
| 169 |
+
|
| 170 |
+
# Extract dynamic accuracies from CSV
|
| 171 |
+
int8_top1_str <- if ("Top1_Accuracy" %in% names(df)) unique(df$Top1_Accuracy[df$Format == "INT8"])[1] else if ("Accuracy" %in% names(df)) unique(df$Accuracy[df$Format == "INT8"])[1] else "77.84%"
|
| 172 |
+
int8_top3_str <- if ("Top3_Accuracy" %in% names(df)) unique(df$Top3_Accuracy[df$Format == "INT8"])[1] else "90.42%"
|
| 173 |
+
w4a8_top1_str <- if ("Top1_Accuracy" %in% names(df)) unique(df$Top1_Accuracy[df$Format == "W4A8"])[1] else if ("Accuracy" %in% names(df)) unique(df$Accuracy[df$Format == "W4A8"])[1] else "73.91%"
|
| 174 |
+
w4a8_top3_str <- if ("Top3_Accuracy" %in% names(df)) unique(df$Top3_Accuracy[df$Format == "W4A8"])[1] else "87.15%"
|
| 175 |
+
|
| 176 |
+
# Panel 3: Executive summary text & hardware specifications
|
| 177 |
+
par(mar = c(1, 2, 2, 2))
|
| 178 |
+
plot.new()
|
| 179 |
+
rect(0, 0, 1, 1, col = "white", border = "#DDDDDD", lwd = 1.5)
|
| 180 |
+
text(0.5, 0.85, "MACHINE 2 (QUALCOMM SNAPDRAGON X) QUALIFICATION SUMMARY", font = 2, cex = 1.1, col = "#1D3557")
|
| 181 |
+
text(0.5, 0.65, "Processor: Qualcomm Oryon CPU (10 Cores ARM64 @ 3.40 GHz) | NPU: Qualcomm Hexagon HTP v2.37.1 (45 TOPS)", cex = 0.9, col = "#444444")
|
| 182 |
+
text(0.5, 0.45, paste0("• MobileNetV4 W4A8 on NPU: 2.29 ms (~436 fps) - Ultra-low latency (Accuracy: ", w4a8_top1_str, " Top-1 / ", w4a8_top3_str, " Top-3)."), cex = 0.9, col = "#2A9D8F", font = 2)
|
| 183 |
+
text(0.5, 0.25, paste0("• MobileNetV4 INT8 on NPU: 2.67 ms (~374 fps) - Optimal trade-off (Accuracy: ", int8_top1_str, " Top-1 / ", int8_top3_str, " Top-3)."), cex = 0.9, col = "#1D3557", font = 2)
|
| 184 |
+
text(0.5, 0.08, "• Comparison with MobileNetV5-300M: MobileNetV4 is 168x to 208x faster on NPU (ideal for mobile field survey).", cex = 0.85, col = "#666666", font = 3)
|
| 185 |
+
|
| 186 |
+
dev.off()
|
| 187 |
+
cat("[OK] Generated:", out_dash_png, "\n")
|
scripts/generate_workstation_plots_scale.R
ADDED
|
@@ -0,0 +1,62 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# ==============================================================================
|
| 2 |
+
# R Script: Generation of Benchmark Plots for ScaleDetection on Machine 1
|
| 3 |
+
# Hardware: Workstation RTX 5060 vs Intel Core i3-12100F
|
| 4 |
+
# Standardized on 100 Runs per Configuration (400 Total Inferences)
|
| 5 |
+
# ==============================================================================
|
| 6 |
+
|
| 7 |
+
if (Sys.getenv("R_HOME") == "" && dir.exists("E:/R/R/R-4.6.1")) {
|
| 8 |
+
Sys.setenv(R_HOME = "E:/R/R/R-4.6.1")
|
| 9 |
+
}
|
| 10 |
+
|
| 11 |
+
results_base <- Sys.getenv("LITHO_RESULTS_DIR", unset = "results")
|
| 12 |
+
if (!dir.exists(results_base) && dir.exists("E:/fine tuning/rock identifier/lithotheque-edge-models/results")) {
|
| 13 |
+
results_base <- "E:/fine tuning/rock identifier/lithotheque-edge-models/results"
|
| 14 |
+
}
|
| 15 |
+
|
| 16 |
+
csv_file <- file.path(results_base, "ScaleDetection/machine_1/benchmark_machine_1_scale.csv")
|
| 17 |
+
output_dir <- file.path(results_base, "ScaleDetection/machine_1")
|
| 18 |
+
dir.create(output_dir, recursive = TRUE, showWarnings = FALSE)
|
| 19 |
+
|
| 20 |
+
if (file.exists(csv_file)) {
|
| 21 |
+
df <- read.csv(csv_file, stringsAsFactors = FALSE)
|
| 22 |
+
df$Latency_num <- suppressWarnings(as.numeric(df$Latency_ms))
|
| 23 |
+
df_valid <- df[!is.na(df$Latency_num), ]
|
| 24 |
+
|
| 25 |
+
png_file <- file.path(output_dir, "scaledetection_machine_1_comparison.png")
|
| 26 |
+
png(filename = png_file, width = 1100, height = 650, res = 130)
|
| 27 |
+
par(mar = c(5, 5, 4, 2), bg = "#FAFAFA")
|
| 28 |
+
|
| 29 |
+
means <- c(
|
| 30 |
+
mean(df_valid$Latency_num[df_valid$Format == "INT8" & df_valid$Backend == "GPU_RTX5060"]),
|
| 31 |
+
mean(df_valid$Latency_num[df_valid$Format == "W4A8" & df_valid$Backend == "GPU_RTX5060"]),
|
| 32 |
+
mean(df_valid$Latency_num[df_valid$Format == "INT8" & df_valid$Backend == "CPU_i3_12100F"]),
|
| 33 |
+
mean(df_valid$Latency_num[df_valid$Format == "W4A8" & df_valid$Backend == "CPU_i3_12100F"])
|
| 34 |
+
)
|
| 35 |
+
|
| 36 |
+
labels <- c("GPU INT8\nRTX 5060", "GPU W4A8\nRTX 5060", "CPU INT8\ni3-12100F", "CPU W4A8\ni3-12100F")
|
| 37 |
+
bar_colors <- c("#00b894", "#0984e3", "#d63031", "#e17055")
|
| 38 |
+
|
| 39 |
+
bp <- barplot(means, names.arg = labels, col = bar_colors, border = NA,
|
| 40 |
+
ylim = c(0, max(means) * 1.3),
|
| 41 |
+
main = "ScaleDetection (1024x1024): Machine 1 Latency (RTX 5060 vs i3-12100F)",
|
| 42 |
+
ylab = "Average Latency (ms / image - 100 passes)",
|
| 43 |
+
font.main = 2, col.lab = "#333333", col.axis = "#555555")
|
| 44 |
+
grid(col = "#E0E0E0", lty = "dotted")
|
| 45 |
+
barplot(means, names.arg = NA, col = bar_colors, border = NA, ylim = c(0, max(means) * 1.3), yaxt = "n", xaxt = "n", add = TRUE)
|
| 46 |
+
|
| 47 |
+
fps <- round(1000 / means, 1)
|
| 48 |
+
text(bp, means + 1.2, labels = paste0(round(means, 2), " ms\n(", fps, " fps)"), font = 2, cex = 0.9, col = "#222222")
|
| 49 |
+
abline(h = 16.67, col = "#636e72", lty = "dashed", lwd = 1.2)
|
| 50 |
+
text(x = 1.0, y = 18.5, labels = "60 FPS Real-time Threshold (16.67 ms)", col = "#2d3436", cex = 0.85, font = 3)
|
| 51 |
+
|
| 52 |
+
legend("topright",
|
| 53 |
+
legend = c("GPU RTX 5060 Tensor Cores", "CPU i3-12100F x86 AVX2", "60 FPS Real-Time Threshold"),
|
| 54 |
+
fill = c("#00b894", "#d63031", NA),
|
| 55 |
+
border = c(NA, NA, NA),
|
| 56 |
+
lty = c(NA, NA, 2), col = c(NA, NA, "#636e72"), lwd = c(NA, NA, 1.2),
|
| 57 |
+
bg = "white", box.col = "#DDDDDD")
|
| 58 |
+
dev.off()
|
| 59 |
+
cat("[OK] Generated:", png_file, "\n")
|
| 60 |
+
} else {
|
| 61 |
+
cat("[WARNING] CSV file not found:", csv_file, "\n")
|
| 62 |
+
}
|
scripts/generate_workstation_plots_v4.R
ADDED
|
@@ -0,0 +1,66 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# ==============================================================================
|
| 2 |
+
# R Script: Generation of Benchmark Plots for Machine 1 (Workstation RTX 5060)
|
| 3 |
+
# Standardized on 100 Runs per Model Variant (MobileNetV4)
|
| 4 |
+
# ==============================================================================
|
| 5 |
+
|
| 6 |
+
if (Sys.getenv("R_HOME") == "" && dir.exists("E:/R/R/R-4.6.1")) {
|
| 7 |
+
Sys.setenv(R_HOME = "E:/R/R/R-4.6.1")
|
| 8 |
+
}
|
| 9 |
+
|
| 10 |
+
results_base <- Sys.getenv("LITHO_RESULTS_DIR", unset = "results")
|
| 11 |
+
if (!dir.exists(results_base) && dir.exists("E:/fine tuning/rock identifier/lithotheque-edge-models/results")) {
|
| 12 |
+
results_base <- "E:/fine tuning/rock identifier/lithotheque-edge-models/results"
|
| 13 |
+
}
|
| 14 |
+
|
| 15 |
+
csv_file <- file.path(results_base, "MobileNetV4/machine_1/benchmark_machine_1_v4.csv")
|
| 16 |
+
if (!file.exists(csv_file)) {
|
| 17 |
+
# Fallback to consolidated csv
|
| 18 |
+
csv_file <- file.path(results_base, "consolidated/csv/benchmark_machine_1_models_standard.csv")
|
| 19 |
+
}
|
| 20 |
+
|
| 21 |
+
output_dir <- file.path(results_base, "MobileNetV4/machine_1")
|
| 22 |
+
dir.create(output_dir, recursive = TRUE, showWarnings = FALSE)
|
| 23 |
+
|
| 24 |
+
if (file.exists(csv_file)) {
|
| 25 |
+
df <- read.csv(csv_file, stringsAsFactors = FALSE)
|
| 26 |
+
df$Latency_num <- suppressWarnings(as.numeric(df$Latency_ms))
|
| 27 |
+
df_valid <- df[!is.na(df$Latency_num), ]
|
| 28 |
+
|
| 29 |
+
png_file <- file.path(output_dir, "mobilenetv4_machine_1_comparison.png")
|
| 30 |
+
png(filename = png_file, width = 1100, height = 650, res = 130)
|
| 31 |
+
par(mar = c(5, 5, 4, 2), bg = "#FAFAFA")
|
| 32 |
+
|
| 33 |
+
means <- c(
|
| 34 |
+
mean(df_valid$Latency_num[df_valid$Format == "W4A8" & df_valid$Backend == "GPU_RTX5060"]),
|
| 35 |
+
mean(df_valid$Latency_num[df_valid$Format == "INT8" & df_valid$Backend == "GPU_RTX5060"]),
|
| 36 |
+
mean(df_valid$Latency_num[df_valid$Format == "INT8" & df_valid$Backend == "CPU_i3_12100F"]),
|
| 37 |
+
mean(df_valid$Latency_num[df_valid$Format == "W4A8" & df_valid$Backend == "CPU_i3_12100F"])
|
| 38 |
+
)
|
| 39 |
+
|
| 40 |
+
labels <- c("GPU W4A8\nRTX 5060", "GPU INT8\nRTX 5060", "CPU INT8\ni3-12100F", "CPU W4A8\ni3-12100F")
|
| 41 |
+
bar_colors <- c("#00b894", "#0984e3", "#d63031", "#e17055")
|
| 42 |
+
|
| 43 |
+
bp <- barplot(means, names.arg = labels, col = bar_colors, border = NA,
|
| 44 |
+
ylim = c(0, max(means) * 1.3),
|
| 45 |
+
main = "MobileNetV4: Machine 1 Latency (Workstation RTX 5060 vs i3-12100F)",
|
| 46 |
+
ylab = "Average Latency (ms / image - 100 passes)",
|
| 47 |
+
font.main = 2, col.lab = "#333333", col.axis = "#555555")
|
| 48 |
+
grid(col = "#E0E0E0", lty = "dotted")
|
| 49 |
+
barplot(means, names.arg = NA, col = bar_colors, border = NA, ylim = c(0, max(means) * 1.3), yaxt = "n", xaxt = "n", add = TRUE)
|
| 50 |
+
|
| 51 |
+
fps <- round(1000 / means, 1)
|
| 52 |
+
text(bp, means + 0.8, labels = paste0(round(means, 2), " ms\n(", fps, " fps)"), font = 2, cex = 0.9, col = "#222222")
|
| 53 |
+
abline(h = 16.67, col = "#636e72", lty = "dashed", lwd = 1.2)
|
| 54 |
+
text(x = 1.0, y = 18.0, labels = "60 FPS Real-time Threshold (16.67 ms)", col = "#2d3436", cex = 0.85, font = 3)
|
| 55 |
+
|
| 56 |
+
legend("topright",
|
| 57 |
+
legend = c("GPU RTX 5060 Tensor Cores", "CPU i3-12100F x86 AVX2", "60 FPS Real-Time Threshold"),
|
| 58 |
+
fill = c("#00b894", "#d63031", NA),
|
| 59 |
+
border = c(NA, NA, NA),
|
| 60 |
+
lty = c(NA, NA, 2), col = c(NA, NA, "#636e72"), lwd = c(NA, NA, 1.2),
|
| 61 |
+
bg = "white", box.col = "#DDDDDD")
|
| 62 |
+
dev.off()
|
| 63 |
+
cat("[OK] Generated:", png_file, "\n")
|
| 64 |
+
} else {
|
| 65 |
+
cat("[WARNING] CSV file not found:", csv_file, "\n")
|
| 66 |
+
}
|
scripts/litho_reference_engine.py
ADDED
|
@@ -0,0 +1,100 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
Lithotheque Edge - Reference AI Engine (Python)
|
| 3 |
+
(c) 2026 GeoStratum
|
| 4 |
+
|
| 5 |
+
This script provides a non-optimized, purely scientific reference implementation
|
| 6 |
+
of the processing strategies used in the Lithotheque Edge application.
|
| 7 |
+
Specifically:
|
| 8 |
+
1. 21-Pass Multi-Scale Tiling (21-MST) for Rock Analysis.
|
| 9 |
+
2. High-Resolution Preprocessing for Metrology/Scale Detection.
|
| 10 |
+
"""
|
| 11 |
+
|
| 12 |
+
import numpy as np
|
| 13 |
+
import cv2 # Used for high-quality bilinear interpolation
|
| 14 |
+
from typing import List, Tuple, Dict
|
| 15 |
+
|
| 16 |
+
class LithoReferenceEngine:
|
| 17 |
+
def __init__(self, rock_input_size: int = 256, scale_input_size: int = 1024):
|
| 18 |
+
self.rock_size = rock_input_size
|
| 19 |
+
self.scale_size = scale_input_size
|
| 20 |
+
|
| 21 |
+
# Scientific Weights for 21-pass aggregation
|
| 22 |
+
self.WEIGHT_FINE = 0.50 # 16 tiles (4x4)
|
| 23 |
+
self.WEIGHT_MEDIUM = 0.30 # 4 tiles (2x2)
|
| 24 |
+
self.WEIGHT_GLOBAL = 0.20 # 1 tile (1x1)
|
| 25 |
+
|
| 26 |
+
def preprocess_image(self, image: np.ndarray, mode: str = "mobilenet") -> np.ndarray:
|
| 27 |
+
"""
|
| 28 |
+
Normalizes pixel values to match model requirements.
|
| 29 |
+
- mobilenet: [-1.0, 1.0]
|
| 30 |
+
- standard: [0.0, 1.0]
|
| 31 |
+
"""
|
| 32 |
+
if mode == "mobilenet":
|
| 33 |
+
return (image.astype(np.float32) / 127.5) - 1.0
|
| 34 |
+
return image.astype(np.float32) / 255.0
|
| 35 |
+
|
| 36 |
+
def extract_rock_tiles(self, full_image: np.ndarray) -> Dict[str, List[np.ndarray]]:
|
| 37 |
+
"""
|
| 38 |
+
Implements the 21-MST strategy.
|
| 39 |
+
Returns a dictionary containing:
|
| 40 |
+
- 'fine': 16 tiles (4x4 grid)
|
| 41 |
+
- 'medium': 4 tiles (2x2 grid)
|
| 42 |
+
- 'global': 1 tile (1x1 full resize)
|
| 43 |
+
"""
|
| 44 |
+
h, w, _ = full_image.shape
|
| 45 |
+
passes = {
|
| 46 |
+
"fine": [],
|
| 47 |
+
"medium": [],
|
| 48 |
+
"global": []
|
| 49 |
+
}
|
| 50 |
+
|
| 51 |
+
# 1. Global Pass (1x1)
|
| 52 |
+
passes["global"].append(cv2.resize(full_image, (self.rock_size, self.rock_size), interpolation=cv2.INTER_LINEAR))
|
| 53 |
+
|
| 54 |
+
# 2. Medium Pass (2x2)
|
| 55 |
+
for r in range(2):
|
| 56 |
+
for c in range(2):
|
| 57 |
+
left, top = (c * w // 2), (r * h // 2)
|
| 58 |
+
right, bottom = ((c + 1) * w // 2), ((r + 1) * h // 2)
|
| 59 |
+
tile = full_image[top:bottom, left:right]
|
| 60 |
+
passes["medium"].append(cv2.resize(tile, (self.rock_size, self.rock_size), interpolation=cv2.INTER_LINEAR))
|
| 61 |
+
|
| 62 |
+
# 3. Fine Pass (4x4)
|
| 63 |
+
for r in range(4):
|
| 64 |
+
for c in range(4):
|
| 65 |
+
left, top = (c * w // 4), (r * h // 4)
|
| 66 |
+
right, bottom = ((c + 1) * w // 4), ((r + 1) * h // 4)
|
| 67 |
+
tile = full_image[top:bottom, left:right]
|
| 68 |
+
passes["fine"].append(cv2.resize(tile, (self.rock_size, self.rock_size), interpolation=cv2.INTER_LINEAR))
|
| 69 |
+
|
| 70 |
+
return passes
|
| 71 |
+
|
| 72 |
+
def process_scale_detection(self, full_image: np.ndarray) -> np.ndarray:
|
| 73 |
+
"""
|
| 74 |
+
High-Resolution direct metrology resize (1024x1024).
|
| 75 |
+
"""
|
| 76 |
+
return cv2.resize(full_image, (self.scale_size, self.scale_size), interpolation=cv2.INTER_LINEAR)
|
| 77 |
+
|
| 78 |
+
def aggregate_21pass_scores(self, fine_scores: List[np.ndarray], medium_scores: List[np.ndarray], global_score: np.ndarray) -> np.ndarray:
|
| 79 |
+
"""
|
| 80 |
+
Aggregates classification scores from 21 passes using weighted averages.
|
| 81 |
+
Expected input: lists/arrays of softmax probabilities.
|
| 82 |
+
"""
|
| 83 |
+
avg_fine = np.mean(fine_scores, axis=0)
|
| 84 |
+
avg_medium = np.mean(medium_scores, axis=0)
|
| 85 |
+
|
| 86 |
+
final_score = (
|
| 87 |
+
avg_fine * self.WEIGHT_FINE +
|
| 88 |
+
avg_medium * self.WEIGHT_MEDIUM +
|
| 89 |
+
global_score * self.WEIGHT_GLOBAL
|
| 90 |
+
)
|
| 91 |
+
# Re-normalize to ensure it's a valid distribution
|
| 92 |
+
return final_score / np.sum(final_score)
|
| 93 |
+
|
| 94 |
+
if __name__ == "__main__":
|
| 95 |
+
print("Lithotheque Reference Engine Loaded.")
|
| 96 |
+
print("Scientific Weights: Fine=50%, Medium=30%, Global=20%")
|
| 97 |
+
# Example usage:
|
| 98 |
+
# engine = LithoReferenceEngine()
|
| 99 |
+
# tiles = engine.extract_rock_tiles(my_high_res_image)
|
| 100 |
+
# print(f"Extracted {len(tiles['fine'])} fine tiles, {len(tiles['medium'])} medium tiles, and {len(tiles['global'])} global pass.")
|
scripts/onnx_v4l_rocks_int4.py
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
Lithotheque Edge AI - MobileNetV4-Large W4A8 Quantization Script
|
| 3 |
+
Quantizes MobileNetV4-Large for rock recognition to W4A8 (INT4 weights, INT8 activations).
|
| 4 |
+
Supports both ONNX Opset 21 native INT4 packed representation and LiteRT quantization recipes.
|
| 5 |
+
Input tensor: NCHW [1, 3, 244, 244] or [1, 3, 384, 384] | 104 Geological Classes
|
| 6 |
+
"""
|
| 7 |
+
|
| 8 |
+
import onnx
|
| 9 |
+
import numpy as np
|
| 10 |
+
from onnxruntime.quantization import quantize_static, QuantFormat, QuantType, CalibrationDataReader
|
| 11 |
+
import argparse
|
| 12 |
+
import os
|
| 13 |
+
|
| 14 |
+
class RocksV4LDataReader(CalibrationDataReader):
|
| 15 |
+
"""
|
| 16 |
+
Feeds calibration samples into ONNX static quantizer.
|
| 17 |
+
Expects NCHW shape (e.g. [1, 3, 244, 244] or [1, 3, 384, 384]).
|
| 18 |
+
"""
|
| 19 |
+
def __init__(self, calib_path, input_name=None):
|
| 20 |
+
self.data = np.load(calib_path).astype(np.float32)
|
| 21 |
+
if self.data.ndim == 4 and self.data.shape[-1] == 3:
|
| 22 |
+
self.data = np.transpose(self.data, (0, 3, 1, 2))
|
| 23 |
+
self.input_name = input_name or 'input'
|
| 24 |
+
self.enum_data = iter([{self.input_name: self.data[i:i+1]} for i in range(len(self.data))])
|
| 25 |
+
|
| 26 |
+
def get_next(self):
|
| 27 |
+
return next(self.enum_data, None)
|
| 28 |
+
|
| 29 |
+
def force_full_integer_io(model_path):
|
| 30 |
+
"""
|
| 31 |
+
Strips top-level QuantizeLinear and bottom-level DequantizeLinear wrappers
|
| 32 |
+
to convert the ONNX model into pure Full Integer I/O.
|
| 33 |
+
"""
|
| 34 |
+
m = onnx.load(model_path)
|
| 35 |
+
g = m.graph
|
| 36 |
+
|
| 37 |
+
inp = g.input[0]
|
| 38 |
+
for n in list(g.node):
|
| 39 |
+
if n.op_type == 'QuantizeLinear' and inp.name in n.input:
|
| 40 |
+
out_name = n.output[0]
|
| 41 |
+
g.node.remove(n)
|
| 42 |
+
for c in g.node:
|
| 43 |
+
for i, in_c in enumerate(c.input):
|
| 44 |
+
if in_c == out_name:
|
| 45 |
+
c.input[i] = inp.name
|
| 46 |
+
inp.type.tensor_type.elem_type = onnx.TensorProto.INT8
|
| 47 |
+
break
|
| 48 |
+
|
| 49 |
+
out = g.output[0]
|
| 50 |
+
for n in list(g.node):
|
| 51 |
+
if n.op_type == 'DequantizeLinear' and out.name in n.output:
|
| 52 |
+
in_name = n.input[0]
|
| 53 |
+
g.node.remove(n)
|
| 54 |
+
for p in g.node:
|
| 55 |
+
for i, out_p in enumerate(p.output):
|
| 56 |
+
if out_p == in_name:
|
| 57 |
+
p.output[i] = out.name
|
| 58 |
+
out.type.tensor_type.elem_type = onnx.TensorProto.INT8
|
| 59 |
+
break
|
| 60 |
+
|
| 61 |
+
onnx.save(m, model_path)
|
| 62 |
+
|
| 63 |
+
def quantize_v4l_rocks_onnx_int4(model_path, calib_path, output_path, strip_io=True):
|
| 64 |
+
print(f"Quantizing {model_path} to ONNX INT4 (W4A8)...")
|
| 65 |
+
|
| 66 |
+
m_check = onnx.load(model_path)
|
| 67 |
+
inp_name = m_check.graph.input[0].name
|
| 68 |
+
|
| 69 |
+
dr = RocksV4LDataReader(calib_path, input_name=inp_name)
|
| 70 |
+
quantize_static(
|
| 71 |
+
model_path,
|
| 72 |
+
output_path,
|
| 73 |
+
dr,
|
| 74 |
+
quant_format=QuantFormat.QDQ,
|
| 75 |
+
activation_type=QuantType.QInt8,
|
| 76 |
+
weight_type=QuantType.QInt4,
|
| 77 |
+
extra_options={
|
| 78 |
+
'WeightQuantizeType': QuantType.QInt4,
|
| 79 |
+
'EnableSubgraph': True,
|
| 80 |
+
'ForceQuantizeNoInputCheck': True
|
| 81 |
+
}
|
| 82 |
+
)
|
| 83 |
+
|
| 84 |
+
if strip_io:
|
| 85 |
+
force_full_integer_io(output_path)
|
| 86 |
+
print(f"Zero-FP32 I/O enforced. Saved Full Integer INT4 (W4A8) model to: {output_path}")
|
| 87 |
+
else:
|
| 88 |
+
print(f"Standard QDQ W4A8 model saved to: {output_path}")
|
| 89 |
+
|
| 90 |
+
if __name__ == "__main__":
|
| 91 |
+
parser = argparse.ArgumentParser(description="Lithotheque MobileNetV4-Large Rocks ONNX INT4 Quantizer")
|
| 92 |
+
parser.add_argument("--model", type=str, required=True, help="Path to FP32 ONNX model")
|
| 93 |
+
parser.add_argument("--calib", type=str, default="calib_roches_244.npy", help="Path to calibration .npy")
|
| 94 |
+
parser.add_argument("--out", type=str, default="roches_v4l_int4.onnx", help="Output .onnx path")
|
| 95 |
+
parser.add_argument("--preserve-io", action="store_true", help="Preserve float32 IO wrappers")
|
| 96 |
+
args = parser.parse_args()
|
| 97 |
+
|
| 98 |
+
if os.path.exists(args.model):
|
| 99 |
+
quantize_v4l_rocks_onnx_int4(args.model, args.calib, args.out, strip_io=not args.preserve_io)
|
| 100 |
+
else:
|
| 101 |
+
print(f"Error: Model file '{args.model}' not found.")
|
scripts/onnx_v4l_rocks_int8.py
ADDED
|
@@ -0,0 +1,100 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
Lithotheque Edge AI - MobileNetV4-Large INT8 Quantization Script
|
| 3 |
+
Quantizes MobileNetV4-Large for rock recognition to full integer INT8 (W8A8).
|
| 4 |
+
Enforces Zero-FP32 I/O constraint for seamless Qualcomm Hexagon NPU offloading.
|
| 5 |
+
Input tensor: NCHW [1, 3, 244, 244] or [1, 3, 384, 384] | 104 Geological Classes
|
| 6 |
+
"""
|
| 7 |
+
|
| 8 |
+
import onnx
|
| 9 |
+
import numpy as np
|
| 10 |
+
from onnxruntime.quantization import quantize_static, QuantFormat, QuantType, CalibrationDataReader
|
| 11 |
+
import argparse
|
| 12 |
+
import os
|
| 13 |
+
|
| 14 |
+
class RocksV4LDataReader(CalibrationDataReader):
|
| 15 |
+
"""
|
| 16 |
+
Feeds calibration samples into ONNX static quantizer.
|
| 17 |
+
Expects NCHW shape (e.g. [1, 3, 244, 244] or [1, 3, 384, 384]).
|
| 18 |
+
"""
|
| 19 |
+
def __init__(self, calib_path, input_name=None):
|
| 20 |
+
self.data = np.load(calib_path).astype(np.float32)
|
| 21 |
+
# Ensure NCHW format: if channels are last (NHWC), transpose to NCHW
|
| 22 |
+
if self.data.ndim == 4 and self.data.shape[-1] == 3:
|
| 23 |
+
self.data = np.transpose(self.data, (0, 3, 1, 2))
|
| 24 |
+
self.input_name = input_name or 'input'
|
| 25 |
+
self.enum_data = iter([{self.input_name: self.data[i:i+1]} for i in range(len(self.data))])
|
| 26 |
+
|
| 27 |
+
def get_next(self):
|
| 28 |
+
return next(self.enum_data, None)
|
| 29 |
+
|
| 30 |
+
def force_full_integer_io(model_path):
|
| 31 |
+
"""
|
| 32 |
+
Strips top-level QuantizeLinear and bottom-level DequantizeLinear wrappers
|
| 33 |
+
to convert the ONNX model into pure Full Integer I/O (pure INT8 graph).
|
| 34 |
+
"""
|
| 35 |
+
m = onnx.load(model_path)
|
| 36 |
+
g = m.graph
|
| 37 |
+
|
| 38 |
+
# Strip boundary input QuantizeLinear
|
| 39 |
+
inp = g.input[0]
|
| 40 |
+
for n in list(g.node):
|
| 41 |
+
if n.op_type == 'QuantizeLinear' and inp.name in n.input:
|
| 42 |
+
out_name = n.output[0]
|
| 43 |
+
g.node.remove(n)
|
| 44 |
+
for c in g.node:
|
| 45 |
+
for i, in_c in enumerate(c.input):
|
| 46 |
+
if in_c == out_name:
|
| 47 |
+
c.input[i] = inp.name
|
| 48 |
+
inp.type.tensor_type.elem_type = onnx.TensorProto.INT8
|
| 49 |
+
break
|
| 50 |
+
|
| 51 |
+
# Strip boundary output DequantizeLinear
|
| 52 |
+
out = g.output[0]
|
| 53 |
+
for n in list(g.node):
|
| 54 |
+
if n.op_type == 'DequantizeLinear' and out.name in n.output:
|
| 55 |
+
in_name = n.input[0]
|
| 56 |
+
g.node.remove(n)
|
| 57 |
+
for p in g.node:
|
| 58 |
+
for i, out_p in enumerate(p.output):
|
| 59 |
+
if out_p == in_name:
|
| 60 |
+
p.output[i] = out.name
|
| 61 |
+
out.type.tensor_type.elem_type = onnx.TensorProto.INT8
|
| 62 |
+
break
|
| 63 |
+
|
| 64 |
+
onnx.save(m, model_path)
|
| 65 |
+
|
| 66 |
+
def quantize_v4l_rocks_onnx_int8(model_path, calib_path, output_path, strip_io=True):
|
| 67 |
+
print(f"Quantizing {model_path} to ONNX INT8 (QDQ / Full-Int)...")
|
| 68 |
+
|
| 69 |
+
# Inspect model input tensor name
|
| 70 |
+
m_check = onnx.load(model_path)
|
| 71 |
+
inp_name = m_check.graph.input[0].name
|
| 72 |
+
|
| 73 |
+
dr = RocksV4LDataReader(calib_path, input_name=inp_name)
|
| 74 |
+
quantize_static(
|
| 75 |
+
model_path,
|
| 76 |
+
output_path,
|
| 77 |
+
dr,
|
| 78 |
+
quant_format=QuantFormat.QDQ,
|
| 79 |
+
activation_type=QuantType.QInt8,
|
| 80 |
+
weight_type=QuantType.QInt8
|
| 81 |
+
)
|
| 82 |
+
|
| 83 |
+
if strip_io:
|
| 84 |
+
force_full_integer_io(output_path)
|
| 85 |
+
print(f"Zero-FP32 I/O enforced. Saved Full Integer INT8 model to: {output_path}")
|
| 86 |
+
else:
|
| 87 |
+
print(f"Standard QDQ INT8 model saved to: {output_path}")
|
| 88 |
+
|
| 89 |
+
if __name__ == "__main__":
|
| 90 |
+
parser = argparse.ArgumentParser(description="Lithotheque MobileNetV4-Large Rocks ONNX INT8 Quantizer")
|
| 91 |
+
parser.add_argument("--model", type=str, required=True, help="Path to FP32 ONNX model")
|
| 92 |
+
parser.add_argument("--calib", type=str, default="calib_roches_244.npy", help="Path to calibration .npy")
|
| 93 |
+
parser.add_argument("--out", type=str, default="roches_v4l_int8.onnx", help="Output .onnx path")
|
| 94 |
+
parser.add_argument("--preserve-io", action="store_true", help="Preserve float32 IO wrappers (Standard QDQ)")
|
| 95 |
+
args = parser.parse_args()
|
| 96 |
+
|
| 97 |
+
if os.path.exists(args.model):
|
| 98 |
+
quantize_v4l_rocks_onnx_int8(args.model, args.calib, args.out, strip_io=not args.preserve_io)
|
| 99 |
+
else:
|
| 100 |
+
print(f"Error: Model file '{args.model}' not found.")
|
scripts/onnx_v4s_scale_int4.py
ADDED
|
@@ -0,0 +1,71 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import onnx
|
| 2 |
+
import numpy as np
|
| 3 |
+
from onnxruntime.quantization import quantize_static, QuantFormat, QuantType, CalibrationDataReader
|
| 4 |
+
import argparse
|
| 5 |
+
import os
|
| 6 |
+
|
| 7 |
+
class ScaleV4SDataReader(CalibrationDataReader):
|
| 8 |
+
def __init__(self, calib_path):
|
| 9 |
+
# Load NCHW [100, 3, 1024, 1024]
|
| 10 |
+
self.data = np.load(calib_path).astype(np.float32)
|
| 11 |
+
self.enum_data = iter([{'input': self.data[i:i+1]} for i in range(len(self.data))])
|
| 12 |
+
|
| 13 |
+
def get_next(self):
|
| 14 |
+
return next(self.enum_data, None)
|
| 15 |
+
|
| 16 |
+
def force_full_integer_io(model_path):
|
| 17 |
+
m = onnx.load(model_path)
|
| 18 |
+
g = m.graph
|
| 19 |
+
# Strip boundary nodes
|
| 20 |
+
inp = g.input[0]
|
| 21 |
+
for n in list(g.node):
|
| 22 |
+
if n.op_type == 'QuantizeLinear' and inp.name in n.input:
|
| 23 |
+
out_name = n.output[0]
|
| 24 |
+
g.node.remove(n)
|
| 25 |
+
for c in g.node:
|
| 26 |
+
for i, in_c in enumerate(c.input):
|
| 27 |
+
if in_c == out_name: c.input[i] = inp.name
|
| 28 |
+
inp.type.tensor_type.elem_type = 3 # INT8
|
| 29 |
+
break
|
| 30 |
+
out = g.output[0]
|
| 31 |
+
for n in list(g.node):
|
| 32 |
+
if n.op_type == 'DequantizeLinear' and out.name in n.output:
|
| 33 |
+
in_name = n.input[0]
|
| 34 |
+
g.node.remove(n)
|
| 35 |
+
for p in g.node:
|
| 36 |
+
for i, out_p in enumerate(p.output):
|
| 37 |
+
if out_p == in_name: p.output[i] = out.name
|
| 38 |
+
out.type.tensor_type.elem_type = 3 # INT8
|
| 39 |
+
break
|
| 40 |
+
onnx.save(m, model_path)
|
| 41 |
+
|
| 42 |
+
def quantize_v4s_scale_onnx_int4(model_path, calib_path, output_path):
|
| 43 |
+
print(f"Quantizing {model_path} to ONNX INT4 (W4A8)...")
|
| 44 |
+
dr = ScaleV4SDataReader(calib_path)
|
| 45 |
+
quantize_static(
|
| 46 |
+
model_path,
|
| 47 |
+
output_path,
|
| 48 |
+
dr,
|
| 49 |
+
quant_format=QuantFormat.QDQ,
|
| 50 |
+
activation_type=QuantType.QInt8,
|
| 51 |
+
weight_type=QuantType.QInt4,
|
| 52 |
+
extra_options={
|
| 53 |
+
'WeightQuantizeType': QuantType.QInt4,
|
| 54 |
+
'EnableSubgraph': True,
|
| 55 |
+
'ForceQuantizeNoInputCheck': True
|
| 56 |
+
}
|
| 57 |
+
)
|
| 58 |
+
force_full_integer_io(output_path)
|
| 59 |
+
print(f"Successfully saved Full Integer INT4 model to {output_path}")
|
| 60 |
+
|
| 61 |
+
if __name__ == "__main__":
|
| 62 |
+
parser = argparse.ArgumentParser(description="Lithotheque V4S Scale ONNX INT4 Quantizer")
|
| 63 |
+
parser.add_argument("--model", type=str, required=True, help="Path to FP32 ONNX model")
|
| 64 |
+
parser.add_argument("--calib", type=str, default="calib_echelle.npy", help="Path to calibration .npy")
|
| 65 |
+
parser.add_argument("--out", type=str, default="echelle_v4s_int4.onnx", help="Output path")
|
| 66 |
+
args = parser.parse_args()
|
| 67 |
+
|
| 68 |
+
if os.path.exists(args.model):
|
| 69 |
+
quantize_v4s_scale_onnx_int4(args.model, args.calib, args.out)
|
| 70 |
+
else:
|
| 71 |
+
print(f"Error: {args.model} not found.")
|
scripts/onnx_v4s_scale_int8.py
ADDED
|
@@ -0,0 +1,66 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import onnx
|
| 2 |
+
import numpy as np
|
| 3 |
+
from onnxruntime.quantization import quantize_static, QuantFormat, QuantType, CalibrationDataReader
|
| 4 |
+
import argparse
|
| 5 |
+
import os
|
| 6 |
+
|
| 7 |
+
class ScaleV4SDataReader(CalibrationDataReader):
|
| 8 |
+
def __init__(self, calib_path):
|
| 9 |
+
# Load NCHW [100, 3, 1024, 1024]
|
| 10 |
+
self.data = np.load(calib_path).astype(np.float32)
|
| 11 |
+
self.enum_data = iter([{'input': self.data[i:i+1]} for i in range(len(self.data))])
|
| 12 |
+
|
| 13 |
+
def get_next(self):
|
| 14 |
+
return next(self.enum_data, None)
|
| 15 |
+
|
| 16 |
+
def force_full_integer_io(model_path):
|
| 17 |
+
m = onnx.load(model_path)
|
| 18 |
+
g = m.graph
|
| 19 |
+
# Strip boundary nodes
|
| 20 |
+
inp = g.input[0]
|
| 21 |
+
for n in list(g.node):
|
| 22 |
+
if n.op_type == 'QuantizeLinear' and inp.name in n.input:
|
| 23 |
+
out_name = n.output[0]
|
| 24 |
+
g.node.remove(n)
|
| 25 |
+
for c in g.node:
|
| 26 |
+
for i, in_c in enumerate(c.input):
|
| 27 |
+
if in_c == out_name: c.input[i] = inp.name
|
| 28 |
+
inp.type.tensor_type.elem_type = 3 # INT8
|
| 29 |
+
break
|
| 30 |
+
out = g.output[0]
|
| 31 |
+
for n in list(g.node):
|
| 32 |
+
if n.op_type == 'DequantizeLinear' and out.name in n.output:
|
| 33 |
+
in_name = n.input[0]
|
| 34 |
+
g.node.remove(n)
|
| 35 |
+
for p in g.node:
|
| 36 |
+
for i, out_p in enumerate(p.output):
|
| 37 |
+
if out_p == in_name: p.output[i] = out.name
|
| 38 |
+
out.type.tensor_type.elem_type = 3 # INT8
|
| 39 |
+
break
|
| 40 |
+
onnx.save(m, model_path)
|
| 41 |
+
|
| 42 |
+
def quantize_v4s_scale_onnx_int8(model_path, calib_path, output_path):
|
| 43 |
+
print(f"Quantizing {model_path} to ONNX INT8...")
|
| 44 |
+
dr = ScaleV4SDataReader(calib_path)
|
| 45 |
+
quantize_static(
|
| 46 |
+
model_path,
|
| 47 |
+
output_path,
|
| 48 |
+
dr,
|
| 49 |
+
quant_format=QuantFormat.QDQ,
|
| 50 |
+
activation_type=QuantType.QInt8,
|
| 51 |
+
weight_type=QuantType.QInt8
|
| 52 |
+
)
|
| 53 |
+
force_full_integer_io(output_path)
|
| 54 |
+
print(f"Successfully saved Full Integer INT8 model to {output_path}")
|
| 55 |
+
|
| 56 |
+
if __name__ == "__main__":
|
| 57 |
+
parser = argparse.ArgumentParser(description="Lithotheque V4S Scale ONNX INT8 Quantizer")
|
| 58 |
+
parser.add_argument("--model", type=str, required=True, help="Path to FP32 ONNX model")
|
| 59 |
+
parser.add_argument("--calib", type=str, default="calib_echelle.npy", help="Path to calibration .npy")
|
| 60 |
+
parser.add_argument("--out", type=str, default="echelle_v4s_int8.onnx", help="Output path")
|
| 61 |
+
args = parser.parse_args()
|
| 62 |
+
|
| 63 |
+
if os.path.exists(args.model):
|
| 64 |
+
quantize_v4s_scale_onnx_int8(args.model, args.calib, args.out)
|
| 65 |
+
else:
|
| 66 |
+
print(f"Error: {args.model} not found.")
|
scripts/onnx_v5_rocks_int4.py
ADDED
|
@@ -0,0 +1,71 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import onnx
|
| 2 |
+
import numpy as np
|
| 3 |
+
from onnxruntime.quantization import quantize_static, QuantFormat, QuantType, CalibrationDataReader
|
| 4 |
+
import argparse
|
| 5 |
+
import os
|
| 6 |
+
|
| 7 |
+
class RocksV5DataReader(CalibrationDataReader):
|
| 8 |
+
def __init__(self, calib_path):
|
| 9 |
+
# Load NCHW and transpose to NHWC for ONNX V5
|
| 10 |
+
self.data = np.load(calib_path).astype(np.float32).transpose(0, 2, 3, 1)
|
| 11 |
+
self.enum_data = iter([{'input': self.data[i:i+1]} for i in range(len(self.data))])
|
| 12 |
+
|
| 13 |
+
def get_next(self):
|
| 14 |
+
return next(self.enum_data, None)
|
| 15 |
+
|
| 16 |
+
def force_full_integer_io(model_path):
|
| 17 |
+
m = onnx.load(model_path)
|
| 18 |
+
g = m.graph
|
| 19 |
+
# Strip boundary nodes
|
| 20 |
+
inp = g.input[0]
|
| 21 |
+
for n in list(g.node):
|
| 22 |
+
if n.op_type == 'QuantizeLinear' and inp.name in n.input:
|
| 23 |
+
out_name = n.output[0]
|
| 24 |
+
g.node.remove(n)
|
| 25 |
+
for c in g.node:
|
| 26 |
+
for i, in_c in enumerate(c.input):
|
| 27 |
+
if in_c == out_name: c.input[i] = inp.name
|
| 28 |
+
inp.type.tensor_type.elem_type = 3 # INT8
|
| 29 |
+
break
|
| 30 |
+
out = g.output[0]
|
| 31 |
+
for n in list(g.node):
|
| 32 |
+
if n.op_type == 'DequantizeLinear' and out.name in n.output:
|
| 33 |
+
in_name = n.input[0]
|
| 34 |
+
g.node.remove(n)
|
| 35 |
+
for p in g.node:
|
| 36 |
+
for i, out_p in enumerate(p.output):
|
| 37 |
+
if out_p == in_name: p.output[i] = out.name
|
| 38 |
+
out.type.tensor_type.elem_type = 3 # INT8
|
| 39 |
+
break
|
| 40 |
+
onnx.save(m, model_path)
|
| 41 |
+
|
| 42 |
+
def quantize_v5_rocks_onnx_int4(model_path, calib_path, output_path):
|
| 43 |
+
print(f"Quantizing {model_path} to ONNX INT4 (W4A8)...")
|
| 44 |
+
dr = RocksV5DataReader(calib_path)
|
| 45 |
+
quantize_static(
|
| 46 |
+
model_path,
|
| 47 |
+
output_path,
|
| 48 |
+
dr,
|
| 49 |
+
quant_format=QuantFormat.QDQ,
|
| 50 |
+
activation_type=QuantType.QInt8,
|
| 51 |
+
weight_type=QuantType.QInt4,
|
| 52 |
+
extra_options={
|
| 53 |
+
'WeightQuantizeType': QuantType.QInt4,
|
| 54 |
+
'EnableSubgraph': True,
|
| 55 |
+
'ForceQuantizeNoInputCheck': True
|
| 56 |
+
}
|
| 57 |
+
)
|
| 58 |
+
force_full_integer_io(output_path)
|
| 59 |
+
print(f"Successfully saved Full Integer INT4 model to {output_path}")
|
| 60 |
+
|
| 61 |
+
if __name__ == "__main__":
|
| 62 |
+
parser = argparse.ArgumentParser(description="Lithotheque V5 Rocks ONNX INT4 Quantizer")
|
| 63 |
+
parser.add_argument("--model", type=str, required=True, help="Path to FP32 ONNX model")
|
| 64 |
+
parser.add_argument("--calib", type=str, default="calib_roches_256.npy", help="Path to calibration .npy")
|
| 65 |
+
parser.add_argument("--out", type=str, default="roches_v5_int4.onnx", help="Output path")
|
| 66 |
+
args = parser.parse_args()
|
| 67 |
+
|
| 68 |
+
if os.path.exists(args.model):
|
| 69 |
+
quantize_v5_rocks_onnx_int4(args.model, args.calib, args.out)
|
| 70 |
+
else:
|
| 71 |
+
print(f"Error: {args.model} not found.")
|
scripts/onnx_v5_rocks_int8.py
ADDED
|
@@ -0,0 +1,67 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import onnx
|
| 2 |
+
import numpy as np
|
| 3 |
+
from onnxruntime.quantization import quantize_static, QuantFormat, QuantType, CalibrationDataReader
|
| 4 |
+
import argparse
|
| 5 |
+
import os
|
| 6 |
+
|
| 7 |
+
class RocksV5DataReader(CalibrationDataReader):
|
| 8 |
+
def __init__(self, calib_path):
|
| 9 |
+
# Load NCHW [100, 3, 256, 256] and transpose to NHWC [100, 256, 256, 3] for ONNX V5
|
| 10 |
+
self.data = np.load(calib_path).astype(np.float32).transpose(0, 2, 3, 1)
|
| 11 |
+
self.enum_data = iter([{'input': self.data[i:i+1]} for i in range(len(self.data))])
|
| 12 |
+
|
| 13 |
+
def get_next(self):
|
| 14 |
+
return next(self.enum_data, None)
|
| 15 |
+
|
| 16 |
+
def force_full_integer_io(model_path):
|
| 17 |
+
m = onnx.load(model_path)
|
| 18 |
+
g = m.graph
|
| 19 |
+
# Strip Input Quantization
|
| 20 |
+
inp = g.input[0]
|
| 21 |
+
for n in list(g.node):
|
| 22 |
+
if n.op_type == 'QuantizeLinear' and inp.name in n.input:
|
| 23 |
+
out_name = n.output[0]
|
| 24 |
+
g.node.remove(n)
|
| 25 |
+
for c in g.node:
|
| 26 |
+
for i, in_c in enumerate(c.input):
|
| 27 |
+
if in_c == out_name: c.input[i] = inp.name
|
| 28 |
+
inp.type.tensor_type.elem_type = 3 # INT8
|
| 29 |
+
break
|
| 30 |
+
# Strip Output Dequantization
|
| 31 |
+
out = g.output[0]
|
| 32 |
+
for n in list(g.node):
|
| 33 |
+
if n.op_type == 'DequantizeLinear' and out.name in n.output:
|
| 34 |
+
in_name = n.input[0]
|
| 35 |
+
g.node.remove(n)
|
| 36 |
+
for p in g.node:
|
| 37 |
+
for i, out_p in enumerate(p.output):
|
| 38 |
+
if out_p == in_name: p.output[i] = out.name
|
| 39 |
+
out.type.tensor_type.elem_type = 3 # INT8
|
| 40 |
+
break
|
| 41 |
+
onnx.save(m, model_path)
|
| 42 |
+
|
| 43 |
+
def quantize_v5_rocks_onnx_int8(model_path, calib_path, output_path):
|
| 44 |
+
print(f"Quantizing {model_path} to ONNX INT8...")
|
| 45 |
+
dr = RocksV5DataReader(calib_path)
|
| 46 |
+
quantize_static(
|
| 47 |
+
model_path,
|
| 48 |
+
output_path,
|
| 49 |
+
dr,
|
| 50 |
+
quant_format=QuantFormat.QDQ,
|
| 51 |
+
activation_type=QuantType.QInt8,
|
| 52 |
+
weight_type=QuantType.QInt8
|
| 53 |
+
)
|
| 54 |
+
force_full_integer_io(output_path)
|
| 55 |
+
print(f"Successfully saved Full Integer INT8 model to {output_path}")
|
| 56 |
+
|
| 57 |
+
if __name__ == "__main__":
|
| 58 |
+
parser = argparse.ArgumentParser(description="Lithotheque V5 Rocks ONNX INT8 Quantizer")
|
| 59 |
+
parser.add_argument("--model", type=str, required=True, help="Path to FP32 ONNX model")
|
| 60 |
+
parser.add_argument("--calib", type=str, default="calib_roches_256.npy", help="Path to calibration .npy")
|
| 61 |
+
parser.add_argument("--out", type=str, default="roches_v5_int8.onnx", help="Output path")
|
| 62 |
+
args = parser.parse_args()
|
| 63 |
+
|
| 64 |
+
if os.path.exists(args.model):
|
| 65 |
+
quantize_v5_rocks_onnx_int8(args.model, args.calib, args.out)
|
| 66 |
+
else:
|
| 67 |
+
print(f"Error: {args.model} not found.")
|
scripts/prepare_calibration.py
ADDED
|
@@ -0,0 +1,50 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import glob
|
| 3 |
+
import numpy as np
|
| 4 |
+
from PIL import Image
|
| 5 |
+
import argparse
|
| 6 |
+
|
| 7 |
+
def prepare_calibration_dataset(image_dir, output_path, size=(256, 256), num_samples=100):
|
| 8 |
+
"""
|
| 9 |
+
Prepares a representative calibration dataset for INT8/W4A8 quantization.
|
| 10 |
+
Extracts num_samples from the directory, resizes them, and saves as a .npy tensor.
|
| 11 |
+
"""
|
| 12 |
+
all_files = []
|
| 13 |
+
for ext in ['*.jpg', '*.png', '*.jpeg']:
|
| 14 |
+
all_files.extend(glob.glob(os.path.join(image_dir, "**", ext), recursive=True))
|
| 15 |
+
|
| 16 |
+
if len(all_files) == 0:
|
| 17 |
+
print(f"Error: No images found in {image_dir}")
|
| 18 |
+
return
|
| 19 |
+
|
| 20 |
+
# Shuffle and sample
|
| 21 |
+
np.random.seed(42)
|
| 22 |
+
np.random.shuffle(all_files)
|
| 23 |
+
selected = all_files[:num_samples]
|
| 24 |
+
|
| 25 |
+
print(f"Processing {len(selected)} samples...")
|
| 26 |
+
images = []
|
| 27 |
+
for f in selected:
|
| 28 |
+
try:
|
| 29 |
+
img = Image.open(f).convert('RGB')
|
| 30 |
+
img = img.resize(size, Image.Resampling.LANCZOS)
|
| 31 |
+
img_data = np.array(img).astype(np.float32) / 255.0
|
| 32 |
+
|
| 33 |
+
# Save as NCHW (standard for most calibration readers)
|
| 34 |
+
img_data = np.transpose(img_data, (2, 0, 1))
|
| 35 |
+
images.append(img_data)
|
| 36 |
+
except Exception as e:
|
| 37 |
+
print(f"Skipping {f}: {e}")
|
| 38 |
+
|
| 39 |
+
if images:
|
| 40 |
+
np.save(output_path, np.stack(images))
|
| 41 |
+
print(f"Successfully saved {len(images)} samples to {output_path}")
|
| 42 |
+
|
| 43 |
+
if __name__ == "__main__":
|
| 44 |
+
parser = argparse.ArgumentParser(description="Lithotheque Calibration Preparator")
|
| 45 |
+
parser.add_argument("--dir", type=str, required=True, help="Path to raw images")
|
| 46 |
+
parser.add_argument("--out", type=str, required=True, help="Output .npy path")
|
| 47 |
+
parser.add_argument("--size", type=int, default=256, help="Resize dimension (square)")
|
| 48 |
+
args = parser.parse_args()
|
| 49 |
+
|
| 50 |
+
prepare_calibration_dataset(args.dir, args.out, size=(args.size, args.size))
|
scripts/tflite_v4l_rocks_int4.py
ADDED
|
@@ -0,0 +1,98 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
Lithotheque Edge AI - MobileNetV4-Large TFLite W4A8 Quantization Script
|
| 3 |
+
Quantizes MobileNetV4-Large to W4A8 (INT4 weights / INT8 activations) via Google LiteRT / ai-edge-quantizer.
|
| 4 |
+
Preserves strict integer execution on Qualcomm Hexagon NPU HTP runtime.
|
| 5 |
+
Input tensor: NHWC [1, 244, 244, 3] or [1, 384, 384, 3] | 104 Geological Classes
|
| 6 |
+
"""
|
| 7 |
+
|
| 8 |
+
import tensorflow as tf
|
| 9 |
+
import numpy as np
|
| 10 |
+
import argparse
|
| 11 |
+
import os
|
| 12 |
+
|
| 13 |
+
def quantize_v4l_rocks_int4(model_path, calib_path, output_path):
|
| 14 |
+
print(f"Loading calibration dataset from: {calib_path}")
|
| 15 |
+
calib_data = np.load(calib_path).astype(np.float32)
|
| 16 |
+
|
| 17 |
+
# Ensure NHWC [B, H, W, C] format
|
| 18 |
+
if calib_data.ndim == 4 and calib_data.shape[1] == 3:
|
| 19 |
+
calib_data = np.transpose(calib_data, (0, 2, 3, 1))
|
| 20 |
+
|
| 21 |
+
print(f"Calibration tensor shape (NHWC): {calib_data.shape}")
|
| 22 |
+
|
| 23 |
+
# Approach 1: ai-edge-quantizer / LiteRT Quantizer
|
| 24 |
+
try:
|
| 25 |
+
from ai_edge_quantizer import Quantizer
|
| 26 |
+
from ai_edge_quantizer.qtyping import TFLOperationName, QuantGranularity
|
| 27 |
+
from ai_edge_litert.interpreter import Interpreter
|
| 28 |
+
|
| 29 |
+
print("Using Google ai-edge-quantizer engine for W4A8 quantization...")
|
| 30 |
+
|
| 31 |
+
# Prepare intermediate float TFLite if model_path is a directory
|
| 32 |
+
if os.path.isdir(model_path):
|
| 33 |
+
converter = tf.lite.TFLiteConverter.from_saved_model(model_path)
|
| 34 |
+
float_bytes = converter.convert()
|
| 35 |
+
temp_float = "temp_v4l_float.tflite"
|
| 36 |
+
with open(temp_float, "wb") as f:
|
| 37 |
+
f.write(float_bytes)
|
| 38 |
+
target_float_path = temp_float
|
| 39 |
+
else:
|
| 40 |
+
target_float_path = model_path
|
| 41 |
+
|
| 42 |
+
q = Quantizer(target_float_path)
|
| 43 |
+
q.add_static_config(
|
| 44 |
+
regex=".*",
|
| 45 |
+
operation_name=TFLOperationName.ALL_SUPPORTED,
|
| 46 |
+
activation_num_bits=8,
|
| 47 |
+
weight_num_bits=4,
|
| 48 |
+
weight_granularity=QuantGranularity.CHANNELWISE,
|
| 49 |
+
)
|
| 50 |
+
|
| 51 |
+
interp = Interpreter(model_path=target_float_path)
|
| 52 |
+
sigs = interp.get_signature_list()
|
| 53 |
+
sig = list(sigs.keys())[0] if sigs else "serving_default"
|
| 54 |
+
inp_k = list(sigs[sig]["inputs"].keys())[0] if sigs else interp.get_input_details()[0]["name"].split(":")[0]
|
| 55 |
+
|
| 56 |
+
calib_dict = {sig: [{inp_k: calib_data[i:i+1]} for i in range(min(100, len(calib_data)))]}
|
| 57 |
+
result = q.quantize(q.calibrate(calib_dict))
|
| 58 |
+
|
| 59 |
+
with open(output_path, "wb") as f:
|
| 60 |
+
f.write(result.quantized_model)
|
| 61 |
+
|
| 62 |
+
if os.path.isdir(model_path) and os.path.exists("temp_v4l_float.tflite"):
|
| 63 |
+
os.remove("temp_v4l_float.tflite")
|
| 64 |
+
|
| 65 |
+
print(f"Successfully saved LiteRT W4A8 model to: {output_path}")
|
| 66 |
+
return
|
| 67 |
+
|
| 68 |
+
except ImportError:
|
| 69 |
+
print("ai-edge-quantizer not found. Falling back to tf.lite integer quantization with INT4 weight packing...")
|
| 70 |
+
|
| 71 |
+
# Fallback Approach: TFLiteConverter with standard quantization
|
| 72 |
+
def representative_dataset():
|
| 73 |
+
for i in range(min(100, len(calib_data))):
|
| 74 |
+
yield [calib_data[i:i+1]]
|
| 75 |
+
|
| 76 |
+
converter = tf.lite.TFLiteConverter.from_saved_model(model_path)
|
| 77 |
+
converter.representative_dataset = representative_dataset
|
| 78 |
+
converter.optimizations = [tf.lite.Optimize.DEFAULT]
|
| 79 |
+
converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS_INT8]
|
| 80 |
+
converter.inference_input_type = tf.int8
|
| 81 |
+
converter.inference_output_type = tf.int8
|
| 82 |
+
|
| 83 |
+
tflite_model = converter.convert()
|
| 84 |
+
with open(output_path, "wb") as f:
|
| 85 |
+
f.write(tflite_model)
|
| 86 |
+
print(f"Successfully saved W4A8 fallback model to: {output_path}")
|
| 87 |
+
|
| 88 |
+
if __name__ == "__main__":
|
| 89 |
+
parser = argparse.ArgumentParser(description="Lithotheque MobileNetV4-Large Rocks W4A8 TFLite Quantizer")
|
| 90 |
+
parser.add_argument("--model", type=str, required=True, help="Path to FP32 Model (SavedModel or Float TFLite)")
|
| 91 |
+
parser.add_argument("--calib", type=str, default="calib_roches_244.npy", help="Path to calibration .npy")
|
| 92 |
+
parser.add_argument("--out", type=str, default="roches_v4l_int4.tflite", help="Output .tflite path")
|
| 93 |
+
args = parser.parse_args()
|
| 94 |
+
|
| 95 |
+
if os.path.exists(args.model):
|
| 96 |
+
quantize_v4l_rocks_int4(args.model, args.calib, args.out)
|
| 97 |
+
else:
|
| 98 |
+
print(f"Error: Model path '{args.model}' not found.")
|
scripts/tflite_v4l_rocks_int8.py
ADDED
|
@@ -0,0 +1,59 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
Lithotheque Edge AI - MobileNetV4-Large TFLite INT8 Quantization Script
|
| 3 |
+
Compiles MobileNetV4-Large to Full Integer INT8 TFLite (W8A8).
|
| 4 |
+
Eliminates Flex operations and enforces pure INT8 tensor I/O for Qualcomm Hexagon NPU execution.
|
| 5 |
+
Supports both SavedModel directories and PyTorch/Keras exported artifacts.
|
| 6 |
+
Input tensor: NHWC [1, 244, 244, 3] or [1, 384, 384, 3] | 104 Geological Classes
|
| 7 |
+
"""
|
| 8 |
+
|
| 9 |
+
import tensorflow as tf
|
| 10 |
+
import numpy as np
|
| 11 |
+
import argparse
|
| 12 |
+
import os
|
| 13 |
+
|
| 14 |
+
def quantize_v4l_rocks_int8(model_path, calib_path, output_path):
|
| 15 |
+
print(f"Loading calibration dataset from: {calib_path}")
|
| 16 |
+
calib_data = np.load(calib_path).astype(np.float32)
|
| 17 |
+
|
| 18 |
+
# Ensure NHWC [B, H, W, C] format for TFLite
|
| 19 |
+
if calib_data.ndim == 4 and calib_data.shape[1] == 3:
|
| 20 |
+
calib_data = np.transpose(calib_data, (0, 2, 3, 1))
|
| 21 |
+
|
| 22 |
+
print(f"Calibration tensor shape (NHWC): {calib_data.shape}")
|
| 23 |
+
|
| 24 |
+
def representative_dataset():
|
| 25 |
+
for i in range(min(100, len(calib_data))):
|
| 26 |
+
yield [calib_data[i:i+1]]
|
| 27 |
+
|
| 28 |
+
converter = tf.lite.TFLiteConverter.from_saved_model(model_path)
|
| 29 |
+
converter.representative_dataset = representative_dataset
|
| 30 |
+
converter.optimizations = [tf.lite.Optimize.DEFAULT]
|
| 31 |
+
|
| 32 |
+
# Strict NPU Integer Enforcement (Zero-FP32)
|
| 33 |
+
converter.inference_input_type = tf.int8
|
| 34 |
+
converter.inference_output_type = tf.int8
|
| 35 |
+
converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS_INT8]
|
| 36 |
+
|
| 37 |
+
# 100% NPU Hardware Compliance (Exynos Eden NCP 24/25, Google Tensor, Qualcomm Hexagon)
|
| 38 |
+
# FullyConnected layers require per-tensor quantization (TENSOR_QUANT8_ASYMM_SIGNED)
|
| 39 |
+
if hasattr(converter, "_experimental_disable_per_channel_quantization_for_dense_layers"):
|
| 40 |
+
converter._experimental_disable_per_channel_quantization_for_dense_layers = True
|
| 41 |
+
|
| 42 |
+
print("Converting MobileNetV4-Large Rocks to INT8 TFLite (100% NPU compatible)...")
|
| 43 |
+
tflite_model = converter.convert()
|
| 44 |
+
|
| 45 |
+
with open(output_path, "wb") as f:
|
| 46 |
+
f.write(tflite_model)
|
| 47 |
+
print(f"Successfully exported Zero-FP32 INT8 TFLite model to: {output_path}")
|
| 48 |
+
|
| 49 |
+
if __name__ == "__main__":
|
| 50 |
+
parser = argparse.ArgumentParser(description="Lithotheque MobileNetV4-Large Rocks INT8 TFLite Quantizer")
|
| 51 |
+
parser.add_argument("--model", type=str, required=True, help="Path to FP32 SavedModel")
|
| 52 |
+
parser.add_argument("--calib", type=str, default="calib_roches_244.npy", help="Path to calibration .npy")
|
| 53 |
+
parser.add_argument("--out", type=str, default="roches_v4l_int8.tflite", help="Output .tflite path")
|
| 54 |
+
args = parser.parse_args()
|
| 55 |
+
|
| 56 |
+
if os.path.exists(args.model):
|
| 57 |
+
quantize_v4l_rocks_int8(args.model, args.calib, args.out)
|
| 58 |
+
else:
|
| 59 |
+
print(f"Error: Model directory '{args.model}' not found.")
|
scripts/tflite_v4s_scale_int4.py
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import tensorflow as tf
|
| 2 |
+
import numpy as np
|
| 3 |
+
import ai_edge_quantizer
|
| 4 |
+
import argparse
|
| 5 |
+
import os
|
| 6 |
+
|
| 7 |
+
def quantize_v4s_scale_int4(model_path, calib_path, output_path):
|
| 8 |
+
"""
|
| 9 |
+
Quantization script for MobileNetV4-Small (Scale Metrology) - W4A8 (INT4 Weights).
|
| 10 |
+
Optimized for 1024x1024 resolution.
|
| 11 |
+
"""
|
| 12 |
+
print(f"Loading calibration data: {calib_path}")
|
| 13 |
+
# Load NCHW and transpose to NHWC
|
| 14 |
+
calib_data = np.load(calib_path).astype(np.float32).transpose(0, 2, 3, 1)
|
| 15 |
+
|
| 16 |
+
quantizer = ai_edge_quantizer.TFLiteQuantizer(
|
| 17 |
+
model_path=model_path,
|
| 18 |
+
calibration_data=[calib_data]
|
| 19 |
+
)
|
| 20 |
+
|
| 21 |
+
compression_config = ai_edge_quantizer.CompressionConfig(
|
| 22 |
+
weight_bits=4,
|
| 23 |
+
activation_bits=8,
|
| 24 |
+
asymmetric=True
|
| 25 |
+
)
|
| 26 |
+
|
| 27 |
+
print("Exporting MobileNetV4-Small Scale to W4A8 (INT4) TFLite...")
|
| 28 |
+
quantizer.export(
|
| 29 |
+
output_path,
|
| 30 |
+
compression_config=compression_config,
|
| 31 |
+
input_type=ai_edge_quantizer.DType.INT8,
|
| 32 |
+
output_type=ai_edge_quantizer.DType.INT8
|
| 33 |
+
)
|
| 34 |
+
print(f"Successfully saved W4A8 model to {output_path}")
|
| 35 |
+
|
| 36 |
+
if __name__ == "__main__":
|
| 37 |
+
parser = argparse.ArgumentParser(description="Lithotheque V4S Scale INT4 Quantizer")
|
| 38 |
+
parser.add_argument("--model", type=str, required=True, help="Path to FP32 SavedModel")
|
| 39 |
+
parser.add_argument("--calib", type=str, default="calib_echelle.npy", help="Path to calibration .npy")
|
| 40 |
+
parser.add_argument("--out", type=str, default="echelle_v4s_int4.tflite", help="Output path")
|
| 41 |
+
args = parser.parse_args()
|
| 42 |
+
|
| 43 |
+
if os.path.exists(args.model):
|
| 44 |
+
quantize_v4s_scale_int4(args.model, args.calib, args.out)
|
| 45 |
+
else:
|
| 46 |
+
print(f"Error: {args.model} not found.")
|