YOLOv4-Paddle

A YOLOv4 reproduction by PaddlePaddle

数据文件准备

数据集已挂载至aistudio项目中，如果需要本地训练可以从这里下载数据集，和标签文件

数据集目录大致如下，可根据实际情况修改

Data
|-- coco
|   |-- annotions
|   |-- images
|      |-- train2017
|      |-- val2017
|      |-- test2017
|   |-- labels
|      |-- train2017
|      |-- val2017
|      |-- train2017.cache(初始解压可删除，训练时会自动生成)
|      |-- val2017.cache(初始解压可删除，训练时会自动生成)
|   |-- test-dev2017.txt
|   |-- val2017.txt
|   |-- train2017.txt

训练

单卡训练

python train.py --batch-size 16 --img 416 416 --data coco.yaml --cfg cfg/yolov4-pacsp.cfg --weights '' --name yolov4-pacsp --notest

多卡训练

python train_multi_gpu.py --batch-size 32 --img 416 416 --data coco.yaml --cfg cfg/yolov4-pacsp.cfg  --weights '' --name yolov4-pacsp --notest

多卡训练项目已提交至脚本任务YOLOv4

多卡训练日志可在此处下载，提取码：0cxk

test-dev数据集验证

python testdev.py --img 416 --conf 0.001 --batch 32 --data coco.yaml --cfg cfg/yolov4-mish-416.cfg --weights weights/yolov4-mish-416.weights

完成后会生成detections_test-dev2017_yolov4_results.json文件，你需要将其压缩为detections_test-dev2017_yolov4_results.zip并在COCO Detection Challenge网站提交

提交完成后等待验证结束，点击View scoring output log即可下载stdout.txt并查看验证情况

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.413
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.622
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.453
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.203
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.450
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.565
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.328
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.527
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.564
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.327
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.620
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.754

验证结果如下所示

Model	Frame	Test Size	APtest	AP50test	AP75test	APStest	APMtest	APLtest	cfg	weights
YOLOv4mish-416	Paddle	416	0.413	0.622	0.453	0.203	0.450	0.565	cfg	weights/r2nw
YOLOv4leaky-416	Paddle	416	0.405	0.616	0.443	0.195	0.441	0.552	cfg	weights/wx7w
YOLOv4	Paddle	416	0.409	0.614	0.447	0.188	0.449	0.572	cfg	weights/lwdy
YOLOv4mish-416	Darknet	416	0.415	0.633	0.447	0.219	0.444	0.553	cfg	weights
YOLOv4leaky-416	Darknet	416	0.407	0.627	0.439	0.214	0.437	0.540	cfg	weights
YOLOv4	Darknet	416	0.412	-	-	-	-	-	cfg	weights

验证所产生的json文件可在此处下载yolov4-mish-416/rmb5，yolov4-leaky-416/nkfo，darknet/lww5

推理

python detect.py --cfg cfg/yolov4-pacsp-x.cfg --weights weights/yolov4-pacsp-x.weights

运行结果将会保存在inference/output文件夹下

GitHub地址

关于作者

姓名	郭权浩
学校	电子科技大学研2020级
研究方向	计算机视觉
主页	Deep Hao的主页
如有错误，请及时留言纠正，非常蟹蟹！
后续会有更多论文复现系列推出，欢迎大家有问题留言交流学习，共同进步成长！