Language: 🇺🇸 🇨🇳

«YOLOv2» reproduced the paper "YOLO9000: Better, Faster, Stronger"

• Train using the VOC07+12 trainval dataset and test using the VOC2007 Test dataset with an input size of 640x640. Given the result as follows

	Original (darknet)	tztztztztz/yolov2.pytorch	zjykzj/YOLOv2(This)	zjykzj/YOLOv2(This)
ARCH	YOLOv2	YOLOv2	YOLOv2	YOLOv2-Fast
GFLOPs	/	/	69.5	48.5
DATASET(TRAIN)	VOC TRAINVAL 2007+2012	VOC TRAINVAL 2007+2012	VOC TRAINVAL 2007+2012	VOC TRAINVAL 2007+2012
DATASET(VAL)	VOC TEST 2007	VOC TEST 2007	VOC TEST 2007	VOC TEST 2007
INPUT_SIZE	416x416	416x416	640x640	640x640
PRETRAINED	TRUE	TRUE	FALSE	FALSE
VOC AP[IoU=0.50:0.95]	/	/	44.3	29.8
VOC AP[IoU=0.50]	76.8	72.7	75.1	62.6

• Train using the COCO train2017 dataset and test using the COCO val2017 dataset with an input size of 640x640. Given the result as follows (Note: The results of the original paper were evaluated on the COCO test-dev2015 dataset)

	Original (darknet)	zjykzj/YOLOv2(This)	zjykzj/YOLOv2(This)
ARCH	YOLOv2	YOLOv2	YOLOv2-Fast
GFLOPs	/	69.7	48.8
DATASET(TRAIN)	/	COCO TRAIN2017	COCO TRAIN2017
DATASET(VAL)	/	COCO VAL2017	COCO VAL2017
INPUT_SIZE	416x416	640x640	640x640
PRETRAINED	TRUE	FALSE	FALSE
COCO AP[IoU=0.50:0.95]	21.6	28.6	20.1
COCO AP[IoU=0.50]	44.0	50.7	41.2

Table of Contents

• Table of Contents
• Latest News
• Background
• Installation
• Usage
  • Train
  • Eval
  • Predict
• Maintainers
• Thanks
• Contributing
• License

Latest News

• [2024/05/04]v1.0.0. Refactoring YOLOv2 project, integrating yolov5 v7.0, reimplementing YOLOv2/YOLOv2-fast and YOLOv2Loss.
• [2023/07/16]v0.3.0. Add ultralytics/yolov5(485da42) transforms.
• [2023/06/28]v0.2.1. Refactor data module.
• [2023/05/21]v0.2.0. Reconstructed loss function and add Darknet53 as a backbone.
• [2023/05/09]v0.1.2. Add COCO dataset result and update VOC dataset training results.
• [2023/05/03]v0.1.1. Fix target transform and update yolov2_voc.cfg and yolov2-tiny_voc.cfg training results for VOC2007 Test.
• [2023/05/02]v0.1.0. Complete YOLOv2 training/evaluation/prediction, while providing the evaluation results of VOC2007 Test.

Background

YOLOv2 has made more innovations on the basis of YOLOv1. For the network, it has created Darknet-19; For the loss function, it adds anchor box settings to help network training with more fine-grained features. Compared with YOLOv1, YOLOv2 is more modern and high-performance.

This repository references many repositories implementations, including tztztztztz/yolov2.pytorch and yjh0410/yolov2-yolov3_PyTorch, as well as zjykzj/YOLOv3.

Note: the latest implementation of YOLOv2 in our warehouse is entirely based on ultralytics/yolov5 v7.0

Installation

pip3 install -r requirements.txt

Or use docker container

docker run -it --runtime nvidia --gpus=all --shm-size=16g -v /etc/localtime:/etc/localtime -v $(pwd):/workdir --workdir=/workdir --name yolov2 ultralytics/yolov5:v7.0

Usage

Train

python3 train.py --data VOC.yaml --weights "" --cfg yolov2_voc.yaml --img 640 --device 0 --yolov2loss
python3 train.py --data VOC.yaml --weights "" --cfg yolov2-fast_voc.yaml --img 640 --device 0 --yolov2loss
python3 train.py --data coco.yaml --weights "" --cfg yolov2.yaml --img 640 --device 0 --yolov2loss
python3 train.py --data coco.yaml --weights "" --cfg yolov2-fast.yaml --img 640 --device 0 --yolov2loss

Eval

# python3 val.py --weights runs/train/voc/exp/weights/best.pt --data VOC.yaml --img 640 --device 0
yolov2_voc summary: 53 layers, 50645053 parameters, 0 gradients, 69.5 GFLOPs
                 Class     Images  Instances          P          R      mAP50   mAP50-95: 100%|██████████| 155/155 00:41
                   all       4952      12032      0.735      0.711      0.751      0.443
Speed: 0.1ms pre-process, 3.1ms inference, 1.3ms NMS per image at shape (32, 3, 640, 640)
# python3 val.py --weights runs/train/voc/exp3/weights/best.pt --data VOC.yaml --img 640 --device 0
yolov2-fast_voc summary: 33 layers, 42367485 parameters, 0 gradients, 48.5 GFLOPs
                 Class     Images  Instances          P          R      mAP50   mAP50-95: 100%|██████████| 155/155 00:37
                   all       4952      12032      0.626      0.612      0.626      0.298
Speed: 0.1ms pre-process, 2.3ms inference, 1.5ms NMS per image at shape (32, 3, 640, 640)
# python3 val.py --weights runs/train/coco/exp/weights/best.pt --data coco.yaml --img 640 --device 0
yolov2 summary: 53 layers, 50952553 parameters, 0 gradients, 69.7 GFLOPs
                 Class     Images  Instances          P          R      mAP50   mAP50-95: 100%|██████████| 157/157 00:57
                   all       5000      36335      0.627       0.48      0.507      0.286
Speed: 0.1ms pre-process, 3.1ms inference, 2.0ms NMS per image at shape (32, 3, 640, 640)
# python3 val.py --weights runs/train/coco/exp2/weights/best.pt --data coco.yaml --img 640 --device 0
yolov2-fast summary: 33 layers, 42674985 parameters, 0 gradients, 48.8 GFLOPs
                 Class     Images  Instances          P          R      mAP50   mAP50-95: 100%|██████████| 157/157 00:53
                   all       5000      36335      0.549      0.402      0.412      0.201
Speed: 0.1ms pre-process, 2.4ms inference, 2.1ms NMS per image at shape (32, 3, 640, 640)

Predict

python3 detect.py --weights runs/yolov2_voc.pt --source ./assets/voc2007-test/

python3 detect.py --weights runs/yolov2_coco.pt --source ./assets/coco/

Maintainers

• zhujian - Initial work - zjykzj

Thanks

• zjykzj/vocdev
• zjykzj/YOLOv3
• zjykzj/anchor-boxes
• ultralytics/yolov5
• AlexeyAB/darknet
• tztztztztz/yolov2.pytorch
• yjh0410/yolov2-yolov3_PyTorch

Contributing

Anyone's participation is welcome! Open an issue or submit PRs.

Small note:

• Git submission specifications should be complied with Conventional Commits
• If versioned, please conform to the Semantic Versioning 2.0.0 specification
• If editing the README, please conform to the standard-readme specification.

License

Apache License 2.0 © 2023 zjykzj