SceneGlue is a scene-aware feature matching framework that overcomes the locality limitation of traditional descriptors by integrating parallel attention for implicit global context modeling and a Visibility Transformer for explicit cross-view visibility estimation. By jointly leveraging implicit and explicit scene-level awareness without requiring scene-level annotations, it significantly improves matching accuracy, robustness, and interpretability across multiple vision tasks. SceneGlue was published in IEEE Transactions on Circuits and Systems for Video Technology, 2026.

https://ieeexplore.ieee.org/document/11483154
https://arxiv.org/abs/2604.13941

• SceneGlue goes beyond traditional point-level matching by introducing scene-aware feature matching, integrating global scene information with local correspondences.
• It combines implicit scene modeling (via parallel self- and cross-attention) with explicit reasoning (via a Visibility Transformer for cross-view visibility estimation).
• Parallel attention mechanism enables more efficient and richer interactions than the sequential attention used in prior methods.
• SceneGlue improves feature representation through multi-scale descriptors and wave-based position encoding, enhancing robustness to scale and geometry variations.

conda env create -f environment.yaml
conda activate sceneglue

We provide the download link to

• the scannet-1500-testset (~1GB).
• the megadepth-1500-testset (~600MB).

You need to setup the testing subsets of ScanNet and MegaDepth first. We create symlinks from the previously downloaded datasets to data/{{dataset}}/test.

# set up symlinks
ln -s /path/to/scannet-1500-testset/* /path/to/LoFTR/data/scannet/test
ln -s /path/to/megadepth-1500-testset/* /path/to/LoFTR/data/megadepth/test

conda activate sceneglue
bash ./scripts/reproduce_test/outdoor.sh

conda activate sceneglue
bash ./scripts/reproduce_test/indoor.sh

@ARTICLE{SceneGlue,
  author={Du, Songlin and Lu, Xiaoyong and Yan, Yaping and Xiao, Guobao and Lu, Xiaobo and Ikenaga, Takeshi},
  journal={IEEE Transactions on Circuits and Systems for Video Technology}, 
  title={SceneGlue: Scene-Aware Transformer for Feature Matching without Scene-Level Annotation}, 
  year={2026},
  doi={10.1109/TCSVT.2026.3684799}
}