This is the joint training model for traffic sign detection and image denoising proposed in our paper titled "CCSPNET-JOINT: Efficient Joint Training Method for Traffic Sign Detection under Extreme Conditions".
The image denoising module of our model utilizes the 4kDehazing model(cite: https://github.com/zzr-idam/4KDehazing.git), while the object detection module incorporates the improved model CCSPNet, which is based on the YOLOv5 baseline, as proposed in our article. This model is a joint training model, and each training session will generate two pth files: "best.pt" for the object detection model and "best_4k.pt" for the image denoising model.
The proposed method and comparisons in this paper were conducted under a unified data augmentation approach. To replicate the experiments, you will need to download the dataset and pre-trained weights and place them in a specific directory. Then, in the terminal, run the command：python train_ccspnet_joint.py --rect
It is worth noting that the joint training model defines a joint loss function calculation formula as loss = alpha * loss1 + beta * loss2, where alpha and beta are hyperparameters. Through extensive experimentation, it has been found that setting alpha = beta = 0.5 yields good results.
The repository includes:
1.CCSPNet model:
CCSPNet-Joint/models/yolov5l-efficientvit-b2-cot.yaml
2.pretrained_pth:
Download link：[https://pan.baidu.com/s/1wfMUxK3Z09R00wus3XzVEA](https://pan.baidu.com/s/1Vo-Xe07KtYYm5TF9Vx4DSQ)
Verification code：1rvo
Content:
ccspnet-joint.pt
our_deblur40.pth
resnet50-0676ba61.pth
checkpoints\efficientViT\b2-r288.pt:
Download link: https://pan.baidu.com/s/1gmXAfND0roMpjCeLO4htlg Verification code：tl7e
3.Dataset:
CCTSDB: https://github.com/csust7zhangjm/CCTSDB.git
Augment method for CCTSDB-AUG: StimulateExtreme.py
4.CCSPNet-Joint/data/ours_aug.yaml
5.train_ccspnet_joint.py
6.detect_joint.py
Please cite our work:
@misc{hong2023ccspnetjoint,
title={CCSPNet-Joint: Efficient Joint Training Method for Traffic Sign Detection Under Extreme Conditions},
author={Haoqin Hong and Yue Zhou and Xiangyu Shu and Xiaofang Hu},
year={2023},
eprint={2309.06902},
archivePrefix={arXiv},
primaryClass={cs.CV}
}