StereoAdapter: Adapting Stereo Depth Estimation to Underwater Scenes

This is the official repository for the paper:

StereoAdapter: Adapting Stereo Depth Estimation to Underwater Scenes

Zhengri Wu*, Yiran Wang*, Yu Wen*, Zeyu Zhang*^†, Biao Wu, and Hao Tang^‡

*Equal contribution. ^†Project lead. ^‡Corresponding author.

Paper | Website | Dataset | Model | HF Paper

Note

💪 This visualizations brief introduce and show the real world deployment of StereoAdapter.

teaser_h264_compressed.mp4

✏️ Citation

If you find our code or paper helpful, please consider starring ⭐ us and citing:

@article{wu2025stereoadapter,
  title={StereoAdapter: Adapting Stereo Depth Estimation to Underwater Scenes},
  author={Wu, Zhengri and Wang, Yiran and Wen, Yu and Zhang, Zeyu and Wu, Biao and Tang, Hao},
  journal={arXiv preprint arXiv:2509.16415},
  year={2025}
}

TODO List

• ✅ Release UW-StereoDepth-40K. (see UW-StereoDepth-40K)
• ⬜️ Upload our paper to arXiv and build project pages.
• ⬜️ Upload the pretrained model.
• ✅ Upload the code.

🏃 Intro StereoAdapter

🔧Run Your StereoAdapter

1. Install & Requirements

conda env create -f environment.yaml
conda activate stereoadapter

2. Download Pertrained Depth Anything V2 Model

mkdir -p Depth-Anything-V2/checkpoints
cd Depth-Anything-V2/checkpoints
wget https://huggingface.co/depth-anything/Depth-Anything-V2-Base/resolve/main/depth_anything_v2_vitb.pth
cd ../..

3. Train StereoAdapter Model

python train_dist_2.py\
  --name MODEL_NAME\
  --exp_opts options/TiO-Depth/train/gru-dav2_codyra-tartanair.yaml\
  --batch_size BATH_SIZE\ 
  --metric_source rawdepth sdepth\ 
  --save_freq SAVE_FREQUENCY\ 
  --visual_freq VISUAL_FREQUENCY\ 
  --is_codyra True\ 
  --step_epochs 20 30

4. Evaluate StereoAdapter

python evaluate.py\
 --exp_opts options/TiO-Depth/train/gru-dav2_codyra-tartanair.yaml\
 --model_path MODEL_PATH