Research map / Video / Video segmentation and tracking
Anchor Diffusion for Unsupervised Video Object Segmentation
Zhao Yang, Qiang Wang, Luca Bertinetto, Weiming Hu, Song Bai, Philip H. S. Torr
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add intra-frame non-local branch (ablation) | DeepLabv3 baseline · DAVIS-2016 val | region similarity J: +0.76 | Improved |
| add intra-frame non-local branch (ablation) | DeepLabv3 baseline · DAVIS-2016 val | contour accuracy F: -0.20 | No clear effect |
| concatenate anchor frame features (ablation) | DeepLabv3 baseline · DAVIS-2016 val | region similarity J: +1.43 | Improved |
| concatenate anchor frame features (ablation) | DeepLabv3 baseline · DAVIS-2016 val | contour accuracy F: +0.18 | No clear effect |
| add anchor diffusion (ablation) | DeepLabv3 baseline · DAVIS-2016 val | region similarity J: +2.02 | Improved |
| add anchor diffusion (ablation) | DeepLabv3 baseline · DAVIS-2016 val | contour accuracy F: +1.20 | Improved |
| add intra-frame branch on top of anchor diffusion (full AD-Net, single scale) (ablation) | Baseline + anchor-diffusion · DAVIS-2016 val | region similarity J: 77.43 -> 78.26 | Improved |
| add intra-frame branch on top of anchor diffusion (full AD-Net, single scale) (ablation) | Baseline + anchor-diffusion · DAVIS-2016 val | contour accuracy F: 76.78 -> 77.11 | Improved |
| use multi-scale and mirrored inputs at inference | AD-Net · DAVIS-2016 val | region similarity J: 78.26 -> 79.4 | Improved |
| add instance pruning post-processing | AD-Net · DAVIS-2016 val | region similarity J: 79.4 -> 81.7 | Improved |
| add instance pruning post-processing | AD-Net · DAVIS-2016 val | contour accuracy F: 78.2 -> 80.5 | Improved |
| sample both training frames randomly instead of using first frame as anchor (ablation) | AD-Net · DAVIS-2016 val | segmentation quality | Hurt |
Closest papers by meaning
- Dense Unsupervised Learning for Video Segmentation (NeurIPS 2021 · similarity 0.90)
- VidSeg: Training-free Video Semantic Segmentation based on Diffusion Models (CVPR 2025 · similarity 0.89)
- Unsupervised Object Segmentation in Video by Efficient Selection of Highly Probable Positive Features (ICCV 2017 · similarity 0.89)
- Learning Spatial-Semantic Features for Robust Video Object Segmentation (ICLR 2025 · similarity 0.88)
- SyncVIS: Synchronized Video Instance Segmentation (NeurIPS 2024 · similarity 0.88)
- Minimizing Labeled, Maximizing Unlabeled: An Image-Driven Approach for Video Instance Segmentation (CVPR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
