Skip to content

Latest commit

 

History

History

object_detection

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
yolo_fastestv2
yolo_fastestv2
 
 
yolov10
yolov10
 
 
README.md
README.md
 
 
README_eng.md
README_eng.md
 
 

README.md

目标检测模型的训练与部署

  1. yolofastestv2
  2. yolov10

模型指标

指标均是在导出的已量化的tflite模型上测量的。没有角标的模型名称表示是使用了迁移学习在预训练模型上进行训练的。"tfs"表示没有使用预训练。"sp"表示使用了空间分离式模型推理。 同时"*"表示该指标来源于stm32ai-modelzoo 。

COCO2017行人

模型名称 实现框架 数据集 输入图像分辨率 mAP MACCs (M) 激活层RAM占用 (KB) ROM占用 (KB) 推理框架
Yolo-FastestV2 sp Pytorch COCO 192x192x3 41.52% 32.47 116.18 385.79 X-CUBE-AI 8.0.1
Yolo-FastestV2 sp tfs Pytorch COCO 192x192x3 39.75% 32.47 116.18 385.79 X-CUBE-AI 8.0.1
Yolo-FastestV2 sp Pytorch COCO 192x192x3 41.52% - 91.76 - TinyEngine
Yolo-FastestV2 sp tfs Pytorch COCO 192x192x3 39.75% - 91.76 - TinyEngine
Yolo-FastestV2 sp Pytorch COCO 256x256x3 49.00% 57.71 214.99 385.79 X-CUBE-AI 8.0.1
Yolo-FastestV2 sp tfs Pytorch COCO 256x256x3 46.87% 57.71 214.99 385.79 X-CUBE-AI 8.0.1
Yolo-FastestV2 sp Pytorch COCO 256x256x3 49.00% - 157.36 - TinyEngine
Yolo-FastestV2 sp tfs Pytorch COCO 256x256x3 46.87% - 157.36 - TinyEngine
YoloV10 sp Pytorch COCO Person 256x256x3 52.45% 57.24 148.86 372.51 X-CUBE-AI 8.0.1
YoloV10 sp tfs Pytorch COCO Person 256x256x3 50.00% 57.24 148.86 372.51 X-CUBE-AI 8.0.1
YoloV10 sp Pytorch COCO Person 256x256x3 52.45% - 144.82 - TinyEngine
YoloV10 sp tfs Pytorch COCO Person 256x256x3 50.00% - 144.82 - TinyEngine
*SSD MobileNet v1 0.25 TensorFlow COCO Person 192x192x3 33.70% 40.48 266.3 438.28 X-CUBE-AI 8.1.0
*SSD MobileNet v1 0.25 TensorFlow COCO Person 256x256x3 46.26% 72.55 456.1 595.66 X-CUBE-AI 8.1.0
*ST Yolo LC v1 tfs TensorFlow COCO Person 192x192x3 31.61% 61.9 166.29 276.73 X-CUBE-AI 9.1.0
*ST Yolo LC v1 tfs TensorFlow COCO Person 256x256x3 40.58% 110.05 278.29 276.73 X-CUBE-AI 9.1.0
*SSD MobileNet v2 0.35 FPN-lite TensorFlow COCO Person 224x224x3 48.67% 167.15 956.82 1007.78 X-CUBE-AI 9.1.0
  • EVAL_IOU = 0.4, NMS_THRESH = 0.5, SCORE_THRESH = 0.001

COCO2017

模型名称 实现框架 输入图像分辨率 mAP MACCs (M) 激活层RAM占用 (KB) ROM占用 (KB) 推理框架
Yolo-FastestV2 sp Pytorch 192x192x3 17.33% 32.47 116.18 385.79 X-CUBE-AI 8.0.1
Yolo-FastestV2 sp tfs Pytorch 192x192x3 11.82% 32.47 116.18 385.79 X-CUBE-AI 8.0.1
Yolo-FastestV2 sp Pytorch 256x256x3 21.11% 57.71 214.99 385.79 X-CUBE-AI 8.0.1
Yolo-FastestV2 sp tfs Pytorch 256x256x3 16.18% 57.71 214.99 385.79 X-CUBE-AI 8.0.1
YoloV10t sp tfs Pytorch 256x256x3 19.21% 57.24 148.86 372.51 X-CUBE-AI 8.0.1
  • EVAL_IOU = 0.4, NMS_THRESH = 0.5, SCORE_THRESH = 0.001

FABD

模型名称 实现框架 输入图像分辨率 mAP MACCs (M) 激活层RAM占用 (KB) ROM占用 (KB) 推理框架
Yolo-FastestV2 sp tfs Pytorch 256x256x3 54.98% 57.71 214.99 385.79 X-CUBE-AI 8.0.1
  • EVAL_IOU = 0.5, NMS_THRESH = 0.4, SCORE_THRESH = 0.01

运行时间

推理时间是使用GCC编译的程序测量出来的时间，ARMCC编译的程序测量出的时间大约多30%。

模型名称 数据格式 图像分辨率 芯片型号 芯片类型 频率 推理时间(ms) 推理框架
Yolo-FastestV2 sp Int8 192x192x3 GD32F470I 1 CPU 240 MHz 500.86 ms X-CUBE-AI 8.0.1
Yolo-FastestV2 sp Int8 192x192x3 GD32F470I 1 CPU 240 MHz 448.24 ms X-CUBE-AI 9.0.0
Yolo-FastestV2 sp Int8 192x192x3 GD32F470I 1 CPU 240 MHz 430.52 ms TinyEngine
Yolo-FastestV2 sp Int8 256x256x3 GD32F470I 1 CPU 240 MHz 863.13 ms X-CUBE-AI 8.0.1
Yolo-FastestV2 sp Int8 256x256x3 GD32F470I 1 CPU 240 MHz 784.14 ms X-CUBE-AI 9.0.0
Yolo-FastestV2 sp Int8 256x256x3 GD32F470I 1 CPU 240 MHz 749.00 ms TinyEngine
Yolo-FastestV2 sp Int8 192x192x3 GD32H759I 1 CPU 600 MHz 143.66 ms X-CUBE-AI 8.0.1
Yolo-FastestV2 sp Int8 192x192x3 GD32H759I 1 CPU 600 MHz 132.73 ms TinyEngine
Yolo-FastestV2 sp Int8 256x256x3 GD32H759I 1 CPU 600 MHz 235.58 ms X-CUBE-AI 8.0.1
Yolo-FastestV2 sp Int8 256x256x3 GD32H759I 1 CPU 600 MHz 221.51 ms X-CUBE-AI 9.0.0
Yolo-FastestV2 sp Int8 256x256x3 GD32H759I 1 CPU 600 MHz 234.13 ms X-CUBE-AI 9.1.0
Yolo-FastestV2 sp Int8 256x256x3 GD32H759I 1 CPU 600 MHz 219.36 ms TinyEngine
YoloV10t sp Int8 256x256x3 GD32F470I 1 CPU 240 MHz 776.14 ms X-CUBE-AI 9.0.0
YoloV10t sp Int8 256x256x3 GD32F470I 1 CPU 240 MHz 707.87 ms TinyEngine
YoloV10t sp Int8 256x256x3 GD32H759I 1 CPU 600 MHz 219.70 ms X-CUBE-AI 8.0.1
YoloV10t sp Int8 256x256x3 GD32H759I 1 CPU 600 MHz 192.57 ms X-CUBE-AI 9.0.0
YoloV10t sp Int8 256x256x3 GD32H759I 1 CPU 600 MHz 208.77 ms X-CUBE-AI 9.1.0
YoloV10t sp Int8 256x256x3 GD32H759I 1 CPU 600 MHz 190.61 ms TinyEngine
*SSD Mobilenet v1 0.25 Int8 192x192x3 STM32H747I 1 CPU 400 MHz 149.22 ms X-CUBE-AI 9.1.0
SSD Mobilenet v1 0.25 Int8 192x192x3 GD32H759I 1 CPU 600 MHz 121.10 ms X-CUBE-AI 9.1.0
*SSD Mobilenet v1 0.25 Int8 256x256x3 STM32H747I 1 CPU 400 MHz 266.4 ms X-CUBE-AI 9.1.0
SSD Mobilenet v1 0.25 Int8 256x256x3 GD32H759I 1 CPU 600 MHz 208.05 ms X-CUBE-AI 9.1.0
*ST Yolo LC v1 Int8 192x192x3 STM32H747I 1 CPU 400 MHz 179.35 ms X-CUBE-AI 9.1.0
*ST Yolo LC v1 Int8 256x256x3 STM32H747I 1 CPU 400 MHz 321.23 ms X-CUBE-AI 9.1.0
*SSD MobileNet v2 0.35 FPN-lite Int8 224x224x3 STM32H747I 1 CPU 400 MHz 675.63 ms X-CUBE-AI 9.1.0