Research map / Video / Video segmentation and tracking
CAVIS: Context-Aware Video Instance Segmentation
Seunghun Lee, Jiwan Seo, Kiljoon Han, Minwoo Choi, Sunghoon Im
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 context-aware contrastive loss L_CTX to the segmenter (ablation) | MinVIS-style segmenter · OVIS val (ResNet-50) | AP: 26.4 -> 29.7 | Improved |
| add PCC loss to the segmenter (ablation) | segmenter · OVIS val (ResNet-50) | AP: 26.4 -> 28.1 | Improved |
| add PCC loss on top of L_CTX (ablation) | segmenter · OVIS val (ResNet-50) | AP: 29.7 -> 30.0 | Improved |
| add context-aware matching in the tracker (ablation) | tracker · OVIS val (ResNet-50) | AP: 35.3 -> 37.6 | Improved |
| align context features in the tracker (ablation) | tracker · OVIS val (ResNet-50) | AP: 33.2 -> 37.6 | Improved |
| use context filter size 9 instead of 3 (ablation) | segmenter · OVIS val (ResNet-50) | AP: 27.3 -> 30.0 | Improved |
| increase context filter size from 9 to 11 (ablation) | segmenter · OVIS val (ResNet-50) | AP: 30.0 -> 29.1 | Hurt |
| use average context filter instead of learnable (ablation) | segmenter · OVIS val (ResNet-50) | AP: 28.7 -> 30.0 | Improved |
| use average filter instead of max (ablation) | segmenter · OVIS val (ResNet-50) | AP: 29.4 -> 30.0 | Improved |
| train with 3 adjacent frames instead of 2 (ablation) | segmenter · OVIS val (ResNet-50) | AP: 29.5 -> 30.0 | Improved |
| train with 4 frames instead of 3 (ablation) | segmenter · OVIS val (ResNet-50) | AP: 30.0 -> 28.7 | Hurt |
| use instance features as value instead of context-aware features (ablation) | tracker context-aware matching · OVIS val (ResNet-50) | AP: 36.8 -> 37.6 | Improved |
Later papers that compared with this method
2 later accepted papers report CAVIS as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- VidEoMT: Your ViT is Secretly Also a Video Segmentation Model (CVPR 2026 · better on 2 of 7 shared results)
- MiNQVIS: Mitigating Noisy Queries for Robust Online Video Instance Segmentation (ECCV 2026 · better on 6 of 6 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- DVIS — DVIS++: Improved Decoupled Framework for Universal Video Segmentation (TPAMI 2025) · better on 12 of 12 shared results
- VISAGE — VISAGE: Video Instance Segmentation with Appearance-Guided Enhancement (ECCV 2024) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MinVIS: A Minimal Video Instance Segmentation Framework without Video-based Training (NeurIPS 2022 · similarity 0.91)
- SyncVIS: Synchronized Video Instance Segmentation (NeurIPS 2024 · similarity 0.89)
- OpenVIS: Open-vocabulary Video Instance Segmentation (AAAI 2025 · similarity 0.87)
- Minimizing Labeled, Maximizing Unlabeled: An Image-Driven Approach for Video Instance Segmentation (CVPR 2025 · similarity 0.87)
- Robust and Consistent Online Video Instance Segmentation via Instance Mask Propagation (AAAI 2025 · similarity 0.86)
- ConViS-Bench: Estimating Video Similarity Through Semantic Concepts (NeurIPS 2025 · similarity 0.86)
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.
