YOLO9tr: Yolo9 with partial self attention

This is the repo for using yolov9 with partial self attention (PSA)
This model was developed to be used in pavement damage detection based on YOLO9s Model.

From paper

YOLO9tr: A Lightweight Model for Pavement Damage Detection Utilizing a Generalized Efficient Layer Aggregation Network and Attention Mechanism Access

Authors

Authors: Dr. Sompote Youwai, Achitaphon Chaiyaphat and Pawarotorn Chaipetch

AI research Group
Department of Civil Engineering
King Mongkut's University of Technology Thonburi
Thailand

Web App

https://huggingface.co/spaces/neng123/pavement_damage

https://yolo9tr.streamlit.app

Deployment

To deploy this project run

git clone https://github.com/Sompote/YOLO9tr
pip install -r  requirements.txt

Reccomend to use weight for YOLO9s as initial training

Train with Single GPU

python train_dual.py --workers 8 --device 0 --batch 4 --data '/workspace/6400 images/data.yaml' --img 640 \
--cfg models/detect/yolov9tr.yaml  --weights '../yolov9s' --name yolov9-tr --hyp hyp.scratch-high.yaml\
 --min-items 0 --epochs 200 --close-mosaic 15

Train with Dual GPU

torchrun  --nproc_per_node 2 --master_port 9527 train_dual.py  \
--workers 8 --device 0,1 --sync-bn --batch 30 --data '/workspace/road damage/data.yaml'  \
--img 640 --cfg models/detect/yolov9tr.yaml --weights '../yolov9s' --name yolov9-c --hyp hyp.scratch-high.yaml \
--min-items 0 --epochs 200 --close-mosaic 15

Evaluation

YOLO9tr.pt

python val_dual.py --data data/coco.yaml --img 640 --batch 32 --conf 0.001\
 --iou 0.7 --device 0 --weights './yolov9tr.pt' \
--save-json --name yolov9_c_640_val

Inference

python detect_dual.py --source './data/images/horses.jpg' --img 640 --device 0 \
--weights './yolov9tr.pt' --name yolov9_c_640_detect

The file format of data can be used the same as YOLOv8 in Roboflow

GPR_YOLO