yjh0410 6a4bb1b9a2 update		1 éve
..
README.md	1b5e49e543 add pretrained	1 éve
build.py	12e96924b0 modify the post-process of YOLOv5	1 éve
loss.py	0744feec35 add YOLOv7-Plus	2 éve
matcher.py	c9d4ce7db1 update	1 éve
yolov5.py	1b5e49e543 add pretrained	1 éve
yolov5_backbone.py	6a4bb1b9a2 update	1 éve
yolov5_basic.py	7cf531e7da add Tracking	2 éve
yolov5_head.py	7cf531e7da add Tracking	2 éve
yolov5_neck.py	1b5e49e543 add pretrained	1 éve
yolov5_pafpn.py	7cf531e7da add Tracking	2 éve

YOLOv5:

For training, we train YOLOv5 series with 300 epochs on COCO.
For data augmentation, we use the large scale jitter (LSJ), Mosaic augmentation and Mixup augmentation, following the setting of YOLOv5.
For optimizer, we use SGD with weight decay 0.0005 and base per image lr 0.01 / 64, following the setting of the official YOLOv5.
For learning rate scheduler, we use linear decay scheduler.
We use decoupled head in our reproduced YOLOv5, which is different from the official YOLOv5'head.

On the other hand, we are trying to use AdamW and larger batch size to train our reproduced YOLOv5. We will update the new results as soon as possible.

Model	Batch	Scale	AP^val 0.5:0.95	AP^val 0.5	FLOPs ^(G)	Params ^(M)	Weight
YOLOv5-N	1xb16	640	29.8	47.1	7.7	2.4	ckpt
YOLOv5-S	1xb16	640	37.8	56.5	27.1	9.0	ckpt
YOLOv5-M	1xb16	640	43.5	62.5	74.3	25.4	ckpt
YOLOv5-L	1xb16	640	46.7	65.5	155.6	54.2	ckpt

For training, we train YOLOv5 series with 300 epochs on COCO.
For data augmentation, we use the large scale jitter (LSJ), Mosaic augmentation and Mixup augmentation, following the setting of YOLOv5.
For optimizer, we use AdamW with weight decay 0.05 and base per image lr 0.001 / 64. We are not good at using SGD.
For learning rate scheduler, we use linear decay scheduler.
We use decoupled head in our reproduced YOLOv5, which is different from the official YOLOv5'head.

Train YOLOv5

Single GPU

Taking training YOLOv5-S on COCO as the example,

python train.py --cuda -d coco --root path/to/coco -m yolov5_s -bs 16 -size 640 --wp_epoch 3 --max_epoch 300 --eval_epoch 10 --no_aug_epoch 20 --ema --fp16 --multi_scale

Multi GPU

Taking training YOLOv5 on COCO as the example,

python -m torch.distributed.run --nproc_per_node=8 train.py --cuda -dist -d coco --root /data/datasets/ -m yolov5_s -bs 128 -size 640 --wp_epoch 3 --max_epoch 300  --eval_epoch 10 --no_aug_epoch 20 --ema --fp16 --sybn --multi_scale --save_folder weights/

Test YOLOv5

Taking testing YOLOv5 on COCO-val as the example,

python test.py --cuda -d coco --root path/to/coco -m yolov5_s --weight path/to/yolov5.pth -size 640 -vt 0.4 --show

Evaluate YOLOv5

Taking evaluating YOLOv5 on COCO-val as the example,

python eval.py --cuda -d coco-val --root path/to/coco -m yolov5_s --weight path/to/yolov5.pth

Demo

Detect with Image

python demo.py --mode image --path_to_img path/to/image_dirs/ --cuda -m yolov5_s --weight path/to/weight -size 640 -vt 0.4 --show

Detect with Video

python demo.py --mode video --path_to_vid path/to/video --cuda -m yolov5_s --weight path/to/weight -size 640 -vt 0.4 --show --gif

Detect with Camera

python demo.py --mode camera --cuda -m yolov5_s --weight path/to/weight -size 640 -vt 0.4 --show --gif

Model	Batch	Scale	AP^val 0.5:0.95	AP^val 0.5	FLOPs ^(G)	Params ^(M)	Weight
YOLOv5-N	8xb16	640
YOLOv5-S	8xb16	640	39.2	57.9	27.3	9.0	ckpt
YOLOv5-M	8xb16	640
YOLOv5-L	8xb16	640
YOLOv5-X	8xb16	640

README.md