🎉🎉🎉NEW WORK! -> [ECCV2024] Rethinking Features-Fused-Pyramid-Neck for Object Detection
- PASCAL VOC 2007+12
- WiderPerson
- SODA10M (for autonomous vehicles)
- DOTA1.0
(We only provide the train/val/test.txt file we used so that you can reproduce our results. The images & labels can be found on the official websites of these datasets.) --- ### An example of comparison on remote sensing images
scaled-yolov4
slim neck scaled-yolov4
(Updated July 14th)
git clone https://github.com/AlanLi1997/slim-neck-by-gsconv.git
cd slim-neck-by-gsconv/gsconv-yolov5
pip install requirements.txt
python train.py --cfg models/sm-yolov5s.yaml
(Updated Aug 17th)
git clone https://github.com/AlanLi1997/slim-neck-by-gsconv.git
cd slim-neck-by-gsconv
pip install requirements.txt
cd gsconv-scaled-yolov4
python train.py --cfg models/sm-yolov4-p5.yaml
MS COCO
| Model | size (pixels) |
mAPval 0.5:0.95 |
mAPval 0.5 |
FPS T4 b1 |
FPS T4 b32 |
params (M) |
FLOPs @640 (G) |
|---|---|---|---|---|---|---|---|
| yolov5n(ultralytics) | 640 | 28.0 | 45.7 | -- | -- | 1.9 | 4.5 |
| GSyolov5n | 640 | 28.4(+0.4) | 47.0(+1.3) | 147 | 207 | 1.8 | 4.0 |
| Model | size (pixels) |
mAPval 0.5:0.95 |
mAPval 0.5 |
FPS A40 b1 |
FPS A40 b32 |
params (M) |
FLOPs @640 (G) |
| yolov5s | 640 | 35.7 | 54.3 | 109 | 297 | 7.2 | 16.4 |
| GSyolov5s | 640 | 36.0(+0.3) | 54.2 | 95 | 312(+15) | 7.0 | 14.5 |
cd gsconv-yolov5
python val.py --data yourdata.yaml --weights sm-yolov5s.pt --task test
cd gsconv-scaled-yolov4
python val.py --data yourdata.yaml --weights sm-yolov4-p5.pt --task test
- https://github.com/ultralytics/yolov5
- https://github.com/AlexeyAB/darknet/tree/yolov4
- https://github.com/WongKinYiu/PyTorch_YOLOv4
- https://github.com/huawei-noah/CV-backbones/tree/master/ghostnet_pytorch
- https://github.com/d-li14/mobilenetv3.pytorch
- https://github.com/megvii-model/ShuffleNet-Series
@article{li2024slim,
title={Slim-neck by GSConv: a lightweight-design for real-time detector architectures},
author={Li, Hulin and Li, Jun and Wei, Hanbing and Liu, Zheng and Zhan, Zhenfei and Ren, Qiliang},
journal={Journal of Real-Time Image Processing},
volume={21},
number={3},
pages={62},
year={2024},
publisher={Springer}
}