{ "scope": { "dataset_split": "val", "input_shape": [1, 3, 640, 640], "batch_size": 1, "precision": "fp16", "dynamic_shapes": false, "warmup_iterations": 50, "benchmark_iterations": 200, "pure_latency_excludes": ["preprocessing", "NMS"], "gpu": "Tesla T4", "tensorrt_version": "10.16.1.11", "cuda_version": "12.8", "torch_version": "2.10.0+cu128", "ultralytics_version": "8.4.115", "peak_gpu_memory_note": "peak_gpu_memory_mib is observed by the PyTorch CUDA allocator; TensorRT execution-context allocation is logged separately at engine load" }, "models": [ { "model": "baseline", "parameters": 3012018, "macs": 4073266400, "pytorch_map50_95": 0.78524, "tensorrt_precision": 0.9675325947571839, "tensorrt_recall": 0.9617830426294427, "tensorrt_map50": 0.9836192902988321, "tensorrt_map50_95": 0.7871632933518187, "pytorch_mean_latency_ms": 8.289, "tensorrt_mean_latency_ms": 1.837303730117128, "tensorrt_median_latency_ms": 1.5013949996500742, "tensorrt_p95_latency_ms": 2.739976000157185, "tensorrt_fps": 544.2758231031568, "tensorrt_engine_size_mib": 7.476772308349609, "pytorch_to_tensorrt_speedup": 4.511502297702066, "trt_baseline_to_model_speedup": 1.0, "validation_preprocess_ms": 1.4128299298863567, "validation_inference_ms": 3.2901282298007573, "validation_postprocess_nms_ms": 2.230188724806794, "peak_gpu_memory_mib": 19.05322265625 }, { "model": "p10_direct", "parameters": 2416871, "macs": 3269461600, "pytorch_map50_95": 0.7773601109, "tensorrt_precision": 0.9639387584067625, "tensorrt_recall": 0.9508054332974109, "tensorrt_map50": 0.9807209311037121, "tensorrt_map50_95": 0.7784167426961883, "pytorch_mean_latency_ms": 10.43257626, "tensorrt_mean_latency_ms": 2.0229106199803937, "tensorrt_median_latency_ms": 1.7276430007768795, "tensorrt_p95_latency_ms": 3.4885230015788693, "tensorrt_fps": 494.3372139742349, "tensorrt_engine_size_mib": 7.627391815185547, "pytorch_to_tensorrt_speedup": 5.157210682942143, "trt_baseline_to_model_speedup": 0.9082476071705706, "validation_preprocess_ms": 1.3793646001977322, "validation_inference_ms": 3.841775689870701, "validation_postprocess_nms_ms": 2.153313165053987, "peak_gpu_memory_mib": 16.8173828125 }, { "model": "p20_direct", "parameters": 1913971, "macs": 2572161600, "pytorch_map50_95": 0.7671004007, "tensorrt_precision": 0.9696520772694063, "tensorrt_recall": 0.9496359835404116, "tensorrt_map50": 0.980323264131055, "tensorrt_map50_95": 0.7693090051241716, "pytorch_mean_latency_ms": 11.716616345, "tensorrt_mean_latency_ms": 1.9325410250530695, "tensorrt_median_latency_ms": 1.761409499522415, "tensorrt_p95_latency_ms": 2.2537910008395556, "tensorrt_fps": 517.453439298935, "tensorrt_engine_size_mib": 7.378376007080078, "pytorch_to_tensorrt_speedup": 6.062803424666367, "trt_baseline_to_model_speedup": 0.9507191341858701, "validation_preprocess_ms": 1.3457183101672854, "validation_inference_ms": 3.8574886749665893, "validation_postprocess_nms_ms": 2.285020630042709, "peak_gpu_memory_mib": 14.8984375 }, { "model": "p30_direct", "parameters": 1452562, "macs": 1961891200, "pytorch_map50_95": 0.7503037353, "tensorrt_precision": 0.9630053238239128, "tensorrt_recall": 0.92858358993507, "tensorrt_map50": 0.9770030959914364, "tensorrt_map50_95": 0.7561045923770547, "pytorch_mean_latency_ms": 9.86312344, "tensorrt_mean_latency_ms": 1.7544926100708835, "tensorrt_median_latency_ms": 1.7208339995704591, "tensorrt_p95_latency_ms": 3.3178800003952347, "tensorrt_fps": 569.965353094077, "tensorrt_engine_size_mib": 5.482402801513672, "pytorch_to_tensorrt_speedup": 5.621638634090068, "trt_baseline_to_model_speedup": 1.0471994692772737, "validation_preprocess_ms": 1.4251806948595913, "validation_inference_ms": 3.515310769980715, "validation_postprocess_nms_ms": 2.0462753048559534, "peak_gpu_memory_mib": 13.15234375 } ] }