Alex Warren, Ke Xu, Xin Tian, Gary K.L. Tam, Benjamin W. Wah, Rynson W.H Lau.

Swansea University, City University of Hong Kong, Huawei Technologies, Chinese University of Hong Kong

[Code and Weights] [Paper] [Swansea University Confra]

To run inference using Motion-in-Depth, you will need to install both a Depth Estimation and a Scene Flow Estimation method. Follow the steps below to set up raft3d (SF) and raft_stereo (Depth Estimation)

networks/
├── raft3d/
└── raft_stereo/

RAFT-3D https://github.com/princeton-vl/RAFT-3D networks/raft3d

RAFT-Stereo https://github.com/princeton-vl/RAFT-Stereo networks/raft_stereo

pip install -r requirements.txt

Inference can be found in train.py

@article{Warren_Xu_Tian_Tam_Wah_Lau_2026, title={Video Mirror Detection with the Motion-in-Depth Cue}, volume={40}, url={https://ojs.aaai.org/index.php/AAAI/article/view/38023}, DOI={10.1609/aaai.v40i13.38023}, year={2026}, month={Mar.}, pages={10512-10520} }