Code of paper Attention-guided Feature Fusion for Small Object Detection
Attention guided feature fusion module for contextual and spatial alignment of adjacent feature maps.
Shallow Feature supplement module for small object feature reinforcement using cross-attention.
For MMYOLO, please see https://mmyolo.readthedocs.io/
Placing files in our repository to the appropriate place like MMYOLO, just overwrite or copy is enough.
Same as MMYOLO, please see 15 minutes to get started with MMYOLO object detection — MMYOLO 0.5.0 documentation and our config file is in configs/att-guided_yolov6s/yolov6_s_myneck.py.
We used a very simple and intuitive way to present the code, with all modules plug and play.
TABLE I. Experimental results on COCO 2017 test-dev
| Method | Backbone | AP | AP50 | AP75 | APS | APM | APL |
|---|---|---|---|---|---|---|---|
| ABFPN | ResNet-50 | 38.6 | 61.3 | - | 24.4 | 42.0 | 49.9 |
| YOLOX-s | Modified CSPNet | 40.5 | 59.7 | 44.2 | 24.1 | 45.2 | 54.0 |
| CL-FPN | ResNet-101 | 41.0 | 62.9 | 44.5 | 23.4 | 44.0 | 52.0 |
| AC-FPN | ResNet-101 | 42.4 | 65.1 | 46.2 | 25.0 | 45.2 | 53.2 |
| PPYOLOE-s | CSPRepResNet | 43.1 | 60.5 | 46.6 | 23.2 | 46.4 | 56.9 |
| YOLOv6-s (baseline) | EfficientRep | 43.5 | 60.4 | 46.8 | 23.7 | 48.9 | 59.9 |
| YOLOv8-s | Modified CSPNet C2f | 44.2 | 61.1 | 47.9 | 25.9 | 49.1 | 60.1 |
| Ours | EfficientRep | 44.3 | 61.8 | 47.4 | 24.6 | 49.6 | 59.9 |
TABLE II. Experimental results on VisDrone2017
| Method | AP | AP50 | AP75 | APS | APM | APL | Epoch |
|---|---|---|---|---|---|---|---|
| YOLOX-s | 17.6 | 33.9 | 16.2 | 9.0 | 27.7 | 45.0 | 50 |
| YOLOv6-s (baseline) | 19.5 | 33.2 | 19.5 | 9.7 | 30.8 | 47.4 | 50 |
| PPYOLOE-s | 20.0 | 34.6 | 20.0 | 10.5 | 31.5 | 51.0 | 50 |
| Zhan et al. | 20.6 | 37.6 | - | - | - | - | 300 |
| YOLOv8-s | 20.9 | 36.8 | 20.7 | 10.7 | 33.2 | 49.7 | 50 |
| FE-YOLOv5 | 21.0 | 37.0 | 20.7 | 13.2 | 29.5 | 39.1 | 300 |
| AMMFN | 24.7 | 48.1 | 22.9 | 17.0 | 43.6 | 60.1 | 300 |
| Ours | 24.1 | 37.5 | 24.7 | 14.2 | 33.8 | 49.2 | 50 |
①
The bold ones mean the top performance.②
Test on RTX3080Ti, after 50 epochs training.jiaxiongyang at tongji dot edu dot cn
If you find this project useful for your research, please use the following BibTeX entry.
@INPROCEEDINGS{10355735,
author={Yang, Jiaxiong and Liu, Xianhui and Liu, Zhuang},
booktitle={2023 IEEE International Conference on Imaging Systems and Techniques (IST)},
title={Attention-guided Feature Fusion for Small Object Detection},
year={2023},
pages={1-6},
doi={10.1109/IST59124.2023.10355735}}