The training and testing experiments are conducted using PyTorch with an NVIDIA A100-SXM of 40 GB Memory.

Note that HDPNet is only tested on Ubuntu OS with the following environments.

• Creating a virtual environment in terminal: conda create -n HDPNet python=3.8.
• Installing necessary packages: pip install -r requirements.txt

• Download the training set (COD-TrainDataset) used for training
• Download the testing sets (COD10K-test + CAMO-test + CHAMELEON + NC4K ) used for testing

• The pretrained model(PVT2) is stored in Google Drive and Baidu Drive (g3ea). After downloading, please change the file path in the corresponding code.
• Run train.sh to train.

Our well-trained model is stored in Google Drive and Baidu Drive (gv9n). After downloading, please change the file path in the corresponding code.

• Evaluate HDPNet: After configuring the test dataset path, run hpvt_eval.sh in the run_slurm folder for evaluation.
• PR-Curves: We provide the code for obtaining PR-Curves through detection results. Please refer to 'PR_Curve.py'.
• Super- and Sub-Classes: To evaluate the performance of each method on COD10K superclasses and subclasses through detection results, please refer to 'class_eval.py'.

The prediction results of our HDPNet are stored on Google Drive. Please check.

Our final results, which perform very well on the COD10K dataset (contains a lot of small objects and detailed labeling of the objects' fine boundaries).

we adopt five kinds of evaluation metrics: S-measure($S_m$), weighted F-measure($F_\beta^\omega$), adaptive F-measure($F_\beta^{adp}$), mean F-measure($F_\beta^{mean}$),max F-measure($F_\beta^{max}$), adaptive E-measure($E_\phi^{apd}$), mean E-measure($E_\phi^{mean}$), max E-measure ($E_\phi^{max}$), and mean absolute error($\mathcal{M}$)

Quantitative results in several typical complex situations, including occlusion, small objects, multiple objects, and object boundaries.

@inproceedings{he2025hdpnet,
  title={HDPNet: Hourglass Vision Transformer with Dual-Path Feature Pyramid for Camouflaged Object Detection},
  author={He, Jinpeng and Liu, Biyuan and Chen, Huaixin},
  booktitle={2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  pages={8638--8647},
  year={2025},
  organization={IEEE}
}