YOLOv4

This is PyTorch implementation of YOLOv4 which is based on ultralytics/yolov3.

development log

Expand

2021-10-31 - support RS loss, aLRP loss, AP loss.
2021-10-30 - support alpha IoU.
2021-10-20 - design resolution calibration methods.
2021-10-15 - support joint detection, instance segmentation, and semantic segmentation. seg-yolo
2021-10-13 - design ratio yolo.
2021-09-22 - pytorch 1.9 compatibility.
2021-09-21 - support DIM.
2021-09-16 - support Dynamic Head.
2021-08-28 - design domain adaptive training.
2021-08-22 - design re-balance models.
2021-08-21 - support simOTA.
2021-08-14 - design approximation-based methods.
2021-07-27 - design new decoders.
2021-07-22 - support 1) decoupled head, 2) anchor-free, and 3) multi positives in yolox.
2021-07-10 - design distribution-based implicit modeling.
2021-07-06 - support outlooker attention. volo
2021-07-06 - design self emsemble training method.
2021-06-23 - design cross multi-stage correlation module.
2021-06-18 - design cross stage cross correlation module.
2021-06-17 - support cross correlation module. ccn
2021-06-17 - support attention modules. cbam saan
2021-04-20 - support swin transformer. swin
2021-03-16 - design new stem layers.
2021-03-13 - design implicit modeling. nn mf lc
2021-01-26 - support vision transformer. tr
2021-01-26 - design mask objectness.
2021-01-25 - design rotate augmentation.
2021-01-23 - design collage augmentation.
2021-01-22 - support VoVNet, VoVNetv2.
2021-01-22 - support EIoU.
2021-01-19 - support instance segmentation. mask-yolo
2021-01-17 - support anchor-free-based methods. center-yolo
2021-01-14 - support joint detection and classification. classify-yolo
2020-01-02 - design new PRN and CSP-based models.
2020-12-22 - support transfer learning.
2020-12-18 - support non-local series self-attention blocks. gc dnl
2020-12-16 - support down-sampling blocks in cspnet paper. down-c down-d
2020-12-03 - support imitation learning.
2020-12-02 - support squeeze and excitation.
2020-11-26 - support multi-class multi-anchor joint detection and embedding.
2020-11-25 - support joint detection and embedding. track-yolo
2020-11-23 - support teacher-student learning.
2020-11-17 - pytorch 1.7 compatibility.
2020-11-06 - support inference with initial weights.
2020-10-21 - fully supported by darknet.
2020-09-18 - design fine-tune methods.
2020-08-29 - support deformable kernel.
2020-08-25 - pytorch 1.6 compatibility.
2020-08-24 - support channel last training/testing.
2020-08-16 - design CSPPRN.
2020-08-15 - design deeper model. csp-p6-mish
2020-08-11 - support HarDNet. hard39-pacsp hard68-pacsp hard85-pacsp
2020-08-10 - add DDP training.
2020-08-06 - support DCN, DCNv2. yolov4-dcn
2020-08-01 - add pytorch hub.
2020-07-31 - support ResNet, ResNeXt, CSPResNet, CSPResNeXt. r50-pacsp x50-pacsp cspr50-pacsp cspx50-pacsp
2020-07-28 - support SAM. yolov4-pacsp-sam
2020-07-24 - update api.
2020-07-23 - support CUDA accelerated Mish activation function.
2020-07-19 - support and training tiny YOLOv4. yolov4-tiny
2020-07-15 - design and training conditional YOLOv4. yolov4-pacsp-conditional
2020-07-13 - support MixUp data augmentation.
2020-07-03 - design new stem layers.
2020-06-16 - support floating16 of GPU inference.
2020-06-14 - convert .pt to .weights for darknet fine-tuning.
2020-06-13 - update multi-scale training strategy.
2020-06-12 - design scaled YOLOv4 follow ultralytics. yolov4-pacsp-s yolov4-pacsp-m yolov4-pacsp-l yolov4-pacsp-x
2020-06-07 - design scaling methods for CSP-based models. yolov4-pacsp-25 yolov4-pacsp-75
2020-06-03 - update COCO2014 to COCO2017.
2020-05-30 - update FPN neck to CSPFPN. yolov4-yocsp yolov4-yocsp-mish
2020-05-24 - update neck of YOLOv4 to CSPPAN. yolov4-pacsp yolov4-pacsp-mish
2020-05-15 - training YOLOv4 with Mish activation function. yolov4-yospp-mish yolov4-paspp-mish
2020-05-08 - design and training YOLOv4 with FPN neck. yolov4-yospp
2020-05-01 - training YOLOv4 with Leaky activation function using PyTorch. yolov4-paspp PAN

Pretrained Models & Comparison

Model	Test Size	AP^test	AP₅₀^test	AP₇₅^test	AP_S^test	AP_M^test	AP_L^test	cfg	weights
YOLOv4	640	50.0%	68.4%	54.7%	30.5%	54.3%	63.3%	cfg	weights

YOLOv4_pacsp-s	640	39.0%	57.8%	42.4%	20.6%	42.6%	50.0%	cfg	weights
YOLOv4_pacsp	640	49.8%	68.4%	54.3%	30.1%	54.0%	63.4%	cfg	weights
YOLOv4_pacsp-x	640	52.2%	70.5%	56.8%	32.7%	56.3%	65.9%	cfg	weights

YOLOv4_pacsp-s-mish	640	40.8%	59.5%	44.3%	22.4%	44.6%	51.8%	cfg	weights
YOLOv4_pacsp-mish	640	50.9%	69.4%	55.5%	31.2%	55.0%	64.7%	cfg	weights
YOLOv4_pacsp-x-mish	640	52.8%	71.1%	57.5%	33.6%	56.9%	66.6%	cfg	weights

Model	Test Size	AP^val	AP₅₀^val	AP₇₅^val	AP_S^val	AP_M^val	AP_L^val	cfg	weights
YOLOv4	640	49.7%	68.2%	54.3%	32.9%	54.8%	63.7%	cfg	weights

YOLOv4_pacsp-s	640	38.9%	57.7%	42.2%	21.9%	43.3%	51.9%	cfg	weights
YOLOv4_pacsp	640	49.4%	68.1%	53.8%	32.7%	54.2%	64.0%	cfg	weights
YOLOv4_pacsp-x	640	51.6%	70.1%	56.2%	35.3%	56.4%	66.9%	cfg	weights

YOLOv4_pacsp-s-mish	640	40.7%	59.5%	44.2%	25.3%	45.1%	53.4%	cfg	weights
YOLOv4_pacsp-mish	640	50.8%	69.4%	55.4%	34.3%	55.5%	65.7%	cfg	weights
YOLOv4_pacsp-x-mish	640	52.6%	71.0%	57.2%	36.4%	57.3%	67.6%	cfg	weights

archive

Model	Test Size	AP^val	AP₅₀^val	AP₇₅^val	AP_S^val	AP_M^val	AP_L^val	cfg	weights
YOLOv4	640	48.4%	67.1%	52.9%	31.7%	53.8%	62.0%	cfg	weights

YOLOv4_pacsp-s	640	37.0%	55.7%	40.0%	20.2%	41.6%	48.4%	cfg	weights
YOLOv4_pacsp	640	47.7%	66.4%	52.0%	32.3%	53.0%	61.7%	cfg	weights
YOLOv4_pacsp-x	640	50.0%	68.3%	54.5%	33.9%	55.4%	63.7%	cfg	weights

YOLOv4_pacsp-s-mish	640	38.8%	57.8%	42.0%	21.6%	43.7%	51.1%	cfg	weights
YOLOv4_pacsp-mish	640	48.8%	67.2%	53.4%	31.5%	54.4%	62.2%	cfg	weights
YOLOv4_pacsp-x-mish	640	51.2%	69.4%	55.9%	35.0%	56.5%	65.0%	cfg	weights

Model	Test Size	AP^val	AP₅₀^val	AP₇₅^val	AP_S^val	AP_M^val	AP_L^val	cfg	weights
YOLOv4	672	47.7%	66.7%	52.1%	30.5%	52.6%	61.4%	cfg	weights

YOLOv4_pacsp-s	672	36.6%	55.5%	39.6%	21.2%	41.1%	47.0%	cfg	weights
YOLOv4_pacsp	672	47.2%	66.2%	51.6%	30.4%	52.3%	60.8%	cfg	weights
YOLOv4_pacsp-x	672	49.3%	68.1%	53.6%	31.8%	54.5%	63.6%	cfg	weights

YOLOv4_pacsp-s-mish	672	38.6%	57.7%	41.8%	22.3%	43.5%	49.3%	cfg	weights
(+BoF)	640	39.9%	59.1%	43.1%	24.4%	45.2%	51.4%		weights
YOLOv4_pacsp-mish	672	48.1%	66.9%	52.3%	30.8%	53.4%	61.7%	cfg	weights
(+BoF)	640	49.3%	68.2%	53.8%	31.9%	54.9%	62.8%		weights
YOLOv4_pacsp-x-mish	672	50.0%	68.5%	54.4%	32.9%	54.9%	64.0%	cfg	weights
(+BoF)	640	51.0%	69.7%	55.5%	33.3%	56.2%	65.5%		weights

Requirements

docker (recommanded):

# create the docker container, you can change the share memory size if you have more.
nvidia-docker run --name yolov4 -it -v your_coco_path/:/coco/ -v your_code_path/:/yolo --shm-size=64g nvcr.io/nvidia/pytorch:20.11-py3

# apt install required packages
apt update
apt install -y zip htop screen libgl1-mesa-glx

# pip install required packages
pip install seaborn thop

# install mish-cuda if you want to use mish activation
# https://github.com/thomasbrandon/mish-cuda
# https://github.com/JunnYu/mish-cuda
cd /
git clone https://github.com/JunnYu/mish-cuda
cd mish-cuda
python setup.py build install

# go to code folder
cd /yolo

local:

pip install -r requirements.txt

※ For running Mish models, please install https://github.com/thomasbrandon/mish-cuda

Training

python train.py --device 0 --batch-size 16 --img 640 640 --data coco.yaml --cfg cfg/yolov4-pacsp.cfg --weights '' --name yolov4-pacsp

Testing

python test.py --img 640 --conf 0.001 --batch 8 --device 0 --data coco.yaml --cfg cfg/yolov4-pacsp.cfg --weights weights/yolov4-pacsp.pt

Citation

@article{bochkovskiy2020yolov4,
  title={{YOLOv4}: Optimal Speed and Accuracy of Object Detection},
  author={Bochkovskiy, Alexey and Wang, Chien-Yao and Liao, Hong-Yuan Mark},
  journal={arXiv preprint arXiv:2004.10934},
  year={2020}
}

@inproceedings{wang2020cspnet,
  title={{CSPNet}: A New Backbone That Can Enhance Learning Capability of {CNN}},
  author={Wang, Chien-Yao and Mark Liao, Hong-Yuan and Wu, Yueh-Hua and Chen, Ping-Yang and Hsieh, Jun-Wei and Yeh, I-Hau},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops},
  pages={390--391},
  year={2020}
}

WongKinYiu/PyTorch_YOLOv4

WongKinYiu

Reviews

Repository Details