[ICASSP 2025] Camouflaged Object Detection via Neural Architecture Search

Xin Li, Keren Fu, Qijun Zhao

✈ Abstract

The core challenge in camouflaged object detection (COD) is identifying objects that blend seamlessly with their surroundings. Existing methods emulate the strategies biological organisms break camouflage by manually constructing modules with expert knowledge from existing segmentation tasks, making it difficult to accurately understand complex and unique camouflage semantics. We are the first to apply neural architecture search (NAS) to COD, introducing an automatic localization and refinement network called ALRNet. It explores a large search space to discover more effective camouflage-specific modules. Specifically, we propose a search-based automatic receptive field block (ARFB) to adaptively excavate hierarchical discriminative cues and decouple features in a multi-branch architecture. Moreover, we introduce an edge-assisted explicit and implicit refinement (EEIR) module, combining explicit priors with implicit search to create a dual-task structure for edge and segmentation knowledge interaction.

✈ Environmental Setups

PyTorch 1.8.0 + CUDA 11.1. Please install corresponding PyTorch and CUDA versions. To create anaconda environment directly, please run flowing commands.

conda create -n ALRNet python=3.8.19
conda activate ALRNet
pip install -r requirements.txt

For the requirements.txt, please click HERE, password: 7qyn.

✈ Dataset and Training

Please download the dataset first and place them in the "COD_dataset" directory. The structure of the "COD_dataset" folder should be as follows:

COD_dataset
└──train_set
    └── edge
    ├── gt
    └── img
└──test_set
    └── CAMO
        ├──im
    	├──gt
    ├── COD10K
    ├── CHAMELEON
    └── NC4K

To train or validate the ALRNet, please run:

python train.py

Download the Res2Net50 backbone weights at HERE, password: 2s5e. Download our ALRNet weights at (HERE, password: 6qcu) into ./checkpoints/Net_epoch_best.pth, modify the para path in train.py, and run infer method in train.py to generate prediction masks. We also provide the prediction masks at HERE, password: rqbs.

✈ Quantitative Results

✈ Visual Results

✈ Citation

If you use ALRNet in your research or wish to refer our work, please use the following BibTeX entry.

@inproceedings{li2025camouflaged,
  title={Camouflaged Object Detection via Neural Architecture Search},
  author={Li, Xin and Fu, Keren and Zhao, Qijun},
  booktitle={ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  pages={1--5},
  year={2025},
  organization={IEEE}
}

✈ The search results of ALRNet: