On the Importance of Backbone to the Adversarial Robustness of Object Detectors

Quick Start

This is the official implementation for On the Importance of Backbone to the Adversarial Robustness of Object Detectors, IEEE TIFS 2025.

We further extend this line of research to defend against physically realizable attacks in our new paper
"PBCAT: Patch-Based Composite Adversarial Training against Physically Realizable Attacks on Object Detection", accepted to ICCV 2025.
🔗 Project page: https://github.com/LixiaoTHU/oddefense-PatchAT

Preparation

conda create -n oddefense python=3.10
conda activate oddefense

conda install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.7 -c pytorch -c nvidia

pip install -U openmim
mim install mmcv-full==1.7.0
pip install mmdet==2.28.0
pip install -r requirements.txt

Download pretrained ResNet-50 backbone: resnet-50 pretrained Download pretrained ConvNeXt-T backbone: convnext-t pretrained

Train and Evaluate

1. Modify Config Files
   Update the following variables in the config files (e.g., frcnn/faster_rcnn_r50_fpn_1x_coco_freeat_all.py):

   • checkpoint_at
   • data_root
   • work_dir

2. Training
   Run the following command to start training:

   bash tools/dist_train.sh [config_file] [num_gpus]

3. Evaluation
   Run the following command to evaluate your model:

   bash tools/dist_test.sh [config_file] [ckpt_path] [num_gpus] --eval bbox

Models

Model	Config File	Checkpoint
Faster-RCNN	faster_rcnn_r50_fpn_1x_coco_freeat_all.py	click to download
FCOS	fcos_r50_caffe_fpn_gn-head_1x_coco_freeat_all.py	click to download
DN-DETR	dn_detr_r50_8x2_12e_coco_freeat_all.py	click to download
Faster-RCNN ConvNeXt	faster_rcnn_convnext_fpn_1x_coco_freeat_all.py	click to download
FCOS ConvNeXt	fcos_convnext_caffe_fpn_gn-head_1x_coco_freeat_all.py	click to download
DN-DETR ConvNeXt	dn_detr_convnext_8x2_12e_coco_freeat_all.py	click to download

Acknowledgement

If you find that our work is helpful to you, please star this project and consider cite:

@article{li2025importance,
  title={On the importance of backbone to the adversarial robustness of object detectors},
  author={Li, Xiao and Chen, Hang and Hu, Xiaolin},
  journal={IEEE Transactions on Information Forensics and Security},
  year={2025},
  publisher={IEEE}
}
@inproceedings{li2025pbcat,
  title={PBCAT: Patch-based composite adversarial training against physically realizable attacks on object detection},
  author={Li, Xiao and Zhu, Yiming and Huang, Yifan and Zhang, Wei and He, Yingzhe and Shi, Jie and Hu, Xiaolin},
  booktitle={{IEEE InternationalConference on Computer Vision},
  year={2025}
}