Release AI-ModelZoo-4.0.0
Browse files
README.md
CHANGED
|
@@ -62,54 +62,61 @@ For an image resolution of NxM and NC classes
|
|
| 62 |
Measures are done with default STM32Cube.AI configuration with enabled input / output allocated option.
|
| 63 |
|
| 64 |
|
| 65 |
-
|
| 66 |
### Reference **NPU** memory footprint based on COCO Person dataset (see Accuracy for details on dataset)
|
| 67 |
-
| Model | Dataset | Format
|
| 68 |
-
|---------------------------------------------------------------------------------------------------------------------|-------------|----------
|
| 69 |
-
| [
|
| 70 |
-
| [
|
| 71 |
-
| [
|
| 72 |
-
|
| 73 |
-
|
|
| 74 |
-
|-
|
| 75 |
-
| [st_yolo_lc_v1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolo_lc_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_lc_v1_192/st_yolo_lc_v1_192_int8.tflite) | COCO-Person | Int8 | 192x192x3 | STM32N6570-DK | NPU/MCU | 1.96 | 510.2 | 10.2.0 | 2.2.0 |
|
| 76 |
-
| [st_yolo_lc_v1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolo_lc_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_lc_v1_224/st_yolo_lc_v1_224_int8.tflite) | COCO-Person | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 2.36 | 423.73 | 10.2.0 | 2.2.0 |
|
| 77 |
-
| [st_yolo_lc_v1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolo_lc_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_lc_v1_224/st_yolo_lc_v1_256_int8.tflite) | COCO-Person | Int8 | 256x256x3 | STM32N6570-DK | NPU/MCU | 3.02 | 331.13 | 10.2.0 | 2.2.0 |
|
| 78 |
|
| 79 |
-
### Reference **MCU** memory footprint based on COCO Person dataset (see Accuracy for details on dataset)
|
| 80 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 81 |
|
| 82 |
-
| Model | Format | Resolution | Series | Activation RAM | Runtime RAM | Weights Flash | Code Flash | Total RAM | Total Flash | STM32Cube.AI version |
|
| 83 |
-
|---------------------------------------------------------------------------------------------------------------------|----------|--------------|----------|------------------|---------------|-----------------|--------------|-------------|---------------|------------------------|
|
| 84 |
-
| [st_yolo_lc_v1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolo_lc_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_lc_v1_192/st_yolo_lc_v1_192_int8.tflite) | Int8 | 192x192x3 | STM32H7 | 166.29 | 8.09 | 276.73 | 52.81 | 174.38 | 329.54 | 10.2.0 |
|
| 85 |
-
| [st_yolo_lc_v1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolo_lc_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_lc_v1_224/st_yolo_lc_v1_224_int8.tflite) | Int8 | 224x224x3 | STM32H7 | 217.29 | 8.09 | 276.73 | 52.82 | 225.38 | 329.55 | 10.2.0 |
|
| 86 |
-
| [st_yolo_lc_v1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolo_lc_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_lc_v1_256/st_yolo_lc_v1_256_int8.tflite) | Int8 | 256x256x3 | STM32H7 | 278.29 | 8.09 | 276.73 | 52.81 | 286.38 | 329.54 | 10.2.0 |
|
| 87 |
|
| 88 |
-
### Reference **MCU**
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 89 |
|
| 90 |
|
| 91 |
-
|
|
|
|
| 92 |
|---------------------------------------------------------------------------------------------------------------------|----------|--------------|------------------|--------------------|-------------|-----------------------|------------------------|
|
| 93 |
-
| [
|
| 94 |
-
| [
|
| 95 |
-
| [
|
| 96 |
|
| 97 |
### Reference **MPU** inference time based on COCO Person dataset (see Accuracy for details on dataset)
|
| 98 |
|
| 99 |
-
| Model
|
| 100 |
-
|-----------
|
| 101 |
-
|
|
| 102 |
-
|
|
| 103 |
-
|
|
| 104 |
-
|
|
| 105 |
-
|
|
| 106 |
-
|
|
| 107 |
-
|
|
| 108 |
-
|
|
| 109 |
-
|
|
| 110 |
|
| 111 |
** **To get the most out of MP25 NPU hardware acceleration, please use per-tensor quantization**
|
| 112 |
|
|
|
|
|
|
|
| 113 |
### AP on COCO Person dataset
|
| 114 |
|
| 115 |
|
|
@@ -117,12 +124,15 @@ Dataset details: [link](https://cocodataset.org/#download) , License [CC BY 4.0]
|
|
| 117 |
|
| 118 |
| Model | Format | Resolution | AP |
|
| 119 |
|-------|--------|------------|----|
|
| 120 |
-
|
|
| 121 |
-
|
|
| 122 |
-
|
|
| 123 |
-
|
|
| 124 |
-
|
|
| 125 |
-
|
|
|
|
|
|
|
|
|
|
|
| 126 |
|
| 127 |
\* EVAL_IOU = 0.5, NMS_THRESH = 0.5, SCORE_THRESH = 0.001, MAX_DETECTIONS = 100
|
| 128 |
|
|
|
|
| 62 |
Measures are done with default STM32Cube.AI configuration with enabled input / output allocated option.
|
| 63 |
|
| 64 |
|
|
|
|
| 65 |
### Reference **NPU** memory footprint based on COCO Person dataset (see Accuracy for details on dataset)
|
| 66 |
+
| Model | Dataset | Format | Resolution | Series | Internal RAM (KiB) | External RAM (KiB) | Weights Flash (KiB) | STEdgeAI Core version |
|
| 67 |
+
|---------------------------------------------------------------------------------------------------------------------|-------------|-------------|--------------|----------|----------------------|----------------------|-----------------------|-------------------------|
|
| 68 |
+
| [st_yololcv1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yololcv1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yololcv1_192/st_yololcv1_192_int8.tflite) | COCO-Person | Int8 | 192x192x3 | STM32N6 | 252 | 0 | 269.44 | 3.0.0 |
|
| 69 |
+
| [st_yololcv1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yololcv1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yololcv1_224/st_yololcv1_224_int8.tflite) | COCO-Person | Int8 | 224x224x3 | STM32N6 | 343 | 0 | 276.19 | 3.0.0 |
|
| 70 |
+
| [st_yololcv1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yololcv1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yololcv1_256/st_yololcv1_256_int8.tflite) | COCO-Person | Int8 | 256x256x3 | STM32N6 | 576 | 0 | 276.19 | 3.0.0 |
|
| 71 |
+
| [st_yololcv1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yololcv1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yololcv1_192/st_yololcv1_192_qdq_w4_74.3%_w8_25.7%_a8_100%_map_33.94.onnx) | COCO-Person | W4A8 | 192x192x3 | STM32N6 | 252 | 0 | 169.42 | 3.0.0 |
|
| 72 |
+
| [st_yololcv1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yololcv1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yololcv1_224/st_yololcv1_224_qdq_w4_50.53%_w8_49.47%_a8_100%_map_34.99.onnx) | COCO-Person | W4A8 | 224x224x3 | STM32N6 | 343 | 0 | 208.17 | 3.0.0 |
|
| 73 |
+
| [st_yololcv1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yololcv1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yololcv1_256/st_yololcv1_256_qdq_w4_50.53%_w8_49.47%_a8_100%_map_36.87.onnx) | COCO-Person | W4A8 | 256x256x3 | STM32N6 | 576 | 0 | 208.19 | 3.0.0 |
|
|
|
|
|
|
|
|
|
|
| 74 |
|
|
|
|
| 75 |
|
| 76 |
+
### Reference **NPU** inference time based on COCO Person dataset (see Accuracy for details on dataset)
|
| 77 |
+
| Model | Dataset | Format | Resolution | Board | Execution Engine | Inference time (ms) | Inf / sec | STEdgeAI Core version |
|
| 78 |
+
|---------------------------------------------------------------------------------------------------------------------|-------------|-------------|--------------|---------------|--------------------|-----------------------|-------------|-------------------------|
|
| 79 |
+
| [st_yololcv1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yololcv1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yololcv1_192/st_yololcv1_192_int8.tflite) | COCO-Person | Int8 | 192x192x3 | STM32N6570-DK | NPU/MCU | 1.90 | 526.32 | 3.0.0 |
|
| 80 |
+
| [st_yololcv1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yololcv1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yololcv1_224/st_yololcv1_224_int8.tflite) | COCO-Person | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 2.26 | 442.48 | 3.0.0 |
|
| 81 |
+
| [st_yololcv1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yololcv1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yololcv1_256/st_yololcv1_256_int8.tflite) | COCO-Person | Int8 | 256x256x3 | STM32N6570-DK | NPU/MCU | 2.90 | 344.83 | 3.0.0 |
|
| 82 |
+
| [st_yololcv1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yololcv1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yololcv1_192/st_yololcv1_192_qdq_w4_74.3%_w8_25.7%_a8_100%_map_33.94.onnx) | COCO-Person | W4A8 | 192x192x3 | STM32N6570-DK | NPU/MCU | 1.85 | 540.54 | 3.0.0 |
|
| 83 |
+
| [st_yololcv1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yololcv1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yololcv1_224/st_yololcv1_224_qdq_w4_50.53%_w8_49.47%_a8_100%_map_34.99.onnx) | COCO-Person | W4A8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 2.26 | 442.48 | 3.0.0 |
|
| 84 |
+
| [st_yololcv1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yololcv1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yololcv1_256/st_yololcv1_256_qdq_w4_50.53%_w8_49.47%_a8_100%_map_36.87.onnx) | COCO-Person | W4A8 | 256x256x3 | STM32N6570-DK | NPU/MCU | 2.82 | 354.61 | 3.0.0 |
|
| 85 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 86 |
|
| 87 |
+
### Reference **MCU** memory footprint based on COCO Person dataset (see Accuracy for details on dataset)
|
| 88 |
+
| Model | Format | Resolution | Series | Activation RAM (KiB) | Runtime RAM (KiB) | Weights Flash (KiB) | Code Flash (KiB) | Total RAM (KiB) | Total Flash (KiB) | STEdgeAI Core version |
|
| 89 |
+
|---------------------------------------------------------------------------------------------------------------------|----------|--------------|----------|------------------------|---------------------|-----------------------|--------------------|--------------------|----------------------|-------------------------|
|
| 90 |
+
| [st_yololcv1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yololcv1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yololcv1_192/st_yololcv1_192_int8.tflite) | Int8 | 192x192x3 | STM32H7 | 166.29 | 0 | 276.73 | 31.15 | 166.29 | 307.88 | 3.0.0 |
|
| 91 |
+
| [st_yololcv1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yololcv1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yololcv1_224/st_yololcv1_224_int8.tflite) | Int8 | 224x224x3 | STM32H7 | 217.29 | 0 | 276.73 | 31.16 | 217.29 | 307.89 | 3.0.0 |
|
| 92 |
+
| [st_yololcv1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yololcv1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yololcv1_256/st_yololcv1_256_int8.tflite) | Int8 | 256x256x3 | STM32H7 | 278.29 | 0 | 276.73 | 31.16 | 278.29 | 307.89 | 3.0.0 |
|
| 93 |
|
| 94 |
|
| 95 |
+
### Reference **MCU** inference time based on COCO Person dataset (see Accuracy for details on dataset)
|
| 96 |
+
| Model | Format | Resolution | Board | Execution Engine | Frequency | Inference time (ms) | STEdgeAI Core version |
|
| 97 |
|---------------------------------------------------------------------------------------------------------------------|----------|--------------|------------------|--------------------|-------------|-----------------------|------------------------|
|
| 98 |
+
| [st_yololcv1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yololcv1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yololcv1_192/st_yololcv1_192_int8.tflite) | Int8 | 192x192x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 186.85 | 3.0.0 |
|
| 99 |
+
| [st_yololcv1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yololcv1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yololcv1_224/st_yololcv1_224_int8.tflite) | Int8 | 224x224x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 255.40 | 3.0.0 |
|
| 100 |
+
| [st_yololcv1](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yololcv1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yololcv1_256/st_yololcv1_256_int8.tflite) | Int8 | 256x256x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 334.78 | 3.0.0 |
|
| 101 |
|
| 102 |
### Reference **MPU** inference time based on COCO Person dataset (see Accuracy for details on dataset)
|
| 103 |
|
| 104 |
+
| Model | Format | Resolution | Quantization | Board | Execution Engine | Frequency | Inference time (ms) | %NPU | %GPU | %CPU | X-LINUX-AI version | Framework |
|
| 105 |
+
|-----------|--------|------------|----------------|-------------------|------------------|-----------|---------------------|-------|-------|------|--------------------|-----------------------|
|
| 106 |
+
| st_yololcv1 | Int8 | 192x192x3 | per-channel** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 5.26 | 93.66 | 6.34 | 0 | v6.1.0 | OpenVX |
|
| 107 |
+
| st_yololcv1 | Int8 | 224x224x3 | per-channel** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 4.90 | 93.55 | 6.45 | 0 | v6.1.0 | OpenVX |
|
| 108 |
+
| st_yololcv1 | Int8 | 256x256x3 | per-channel** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 6.43 | 94.18 | 5.82 | 0 | v6.1.0 | OpenVX |
|
| 109 |
+
| st_yololcv1 | Int8 | 192x192x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 51.42 | NA | NA | 100 | v6.1.0 | TensorFlowLite 2.18.0 |
|
| 110 |
+
| st_yololcv1 | Int8 | 224x224x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 72.44 | NA | NA | 100 | v6.1.0 | TensorFlowLite 2.18.0 |
|
| 111 |
+
| st_yololcv1 | Int8 | 256x256x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 88.55 | NA | NA | 100 | v6.1.0 | TensorFlowLite 2.18.0 |
|
| 112 |
+
| st_yololcv1 | Int8 | 192x192x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 79.26 | NA | NA | 100 | v6.1.0 | TensorFlowLite 2.18.0 |
|
| 113 |
+
| st_yololcv1 | Int8 | 224x224x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 106.30 | NA | NA | 100 | v6.1.0 | TensorFlowLite 2.18.0 |
|
| 114 |
+
| st_yololcv1 | Int8 | 256x256x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 140.87 | NA | NA | 100 | v6.1.0 | TensorFlowLite 2.18.0 |
|
| 115 |
|
| 116 |
** **To get the most out of MP25 NPU hardware acceleration, please use per-tensor quantization**
|
| 117 |
|
| 118 |
+
** **Note:** On STM32MP2 devices, per-channel quantized models are internally converted to per-tensor quantization by the compiler using an entropy-based method. This may introduce a slight loss in accuracy compared to the original per-channel models.
|
| 119 |
+
|
| 120 |
### AP on COCO Person dataset
|
| 121 |
|
| 122 |
|
|
|
|
| 124 |
|
| 125 |
| Model | Format | Resolution | AP |
|
| 126 |
|-------|--------|------------|----|
|
| 127 |
+
| st_yololcv1 | Int8 | 192x192x3 | 34.7% |
|
| 128 |
+
| st_yololcv1 | Float | 192x192x3 | 34.9 % |
|
| 129 |
+
| st_yololcv1 | w4w8 | 192x192x3 | 33.94 % |
|
| 130 |
+
| st_yololcv1 | Int8 | 224x224x3 | 35.5 % |
|
| 131 |
+
| st_yololcv1 | Float | 224x224x3 | 35.8 % |
|
| 132 |
+
| st_yololcv1 | w4w8 | 224x224x3 | 34.99 % |
|
| 133 |
+
| st_yololcv1 | Int8 | 256x256x3 | 37.2 % |
|
| 134 |
+
| st_yololcv1 | Float | 256x256x3 | 37.3 % |
|
| 135 |
+
| st_yololcv1 | w4w8 | 256x256x3 | 36.87 % |
|
| 136 |
|
| 137 |
\* EVAL_IOU = 0.5, NMS_THRESH = 0.5, SCORE_THRESH = 0.001, MAX_DETECTIONS = 100
|
| 138 |
|