Rethinking Detecting Salient and Camouflaged Objects in Unconstrained Scenes

Zhangjun Zhou ¹,  Yiping Li ¹,  Chunlin Zhong ¹,  Jianuo Huang ¹,  Jialun Pei ²,  Hua Li ³,  He Tang ^1*

¹ Huazhong University of Science and Technology  ² The Chinese University of Hong Kong  ³ Hainan University 

               

This repo is the official implementation of "Rethinking Detecting Salient and Camouflaged Objects in Unconstrained Scenes" (ICCV 2025).

Contact: [email protected]; [email protected]

Environment preparation

Requirements

• Please refer to the link: SAM.
• You may need to install Apex using pip.

Dataset preparation 🔥

Download the datasets and annotation files

• USC12K: Baidu/ Google.
• USC12K class files: Google

Register datasets

1. Download the datasets and put them in the same folder. To match the folder name in the dataset mappers, you'd better not change the folder names, its structure may be:

    DATASET_ROOT/
    ├── VOC-USC12K
       ├── ImageSets
          ├── Segmentation
              ├── Scene-A.txt
              ├── Scene-B.txt
              ├── Scene-C.txt
              ├── Scene-D.txt
              ├── train.txt
              ├── val.txt
       ├── JPEGImages
       ├── SegmentationClass

Pre-trained models :

• Download the pre-training weights of SAM ViT-H: sam_vit_h_4b8939.pth.
• Download the pre-trained weights on USC12K: Baidu/ Google.

Visualization results ⚡

The visual results of SOTAs on USC12K test set.

• Results on the Overall Scene test set: Baidu/ Google.

Usage

Train&Test

• To train our USCNet on single GPU by following command,the trained models will be saved in savePath folder. You can modify datapath if you want to run your own datases.

bash train.sh

• To test and evaluate our USCNet on USC12K:

bash test.sh

A PPT for introduction and future exploration

Only chinese PPT now: MIR-Shared PPT.

To watch the video: Bilibili link.

Acknowledgement

Additional thanks to the following contributors to this project: Huaiyu Chen, Weiyi Cui, Mingxin Yang, Mengzhe Cui, Fei Liu, Yan Xu, Haopeng Fang, and Xiaokai Zhang from the School of Software Engineering, Huazhong University of Science and Technology.

Citation

If this helps you, please cite this work:

@inproceedings{zhou2025rethinking,
  title={Rethinking Detecting Salient and Camouflaged Objects in Unconstrained Scenes},
  author={Zhou, Zhangjun and Li, Yiping and Zhong, Chunlin and Huang, Jianuo and Pei, Jialun and Li, Hua and Tang, He},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={22372--22382},
  year={2025}
}
}