Research map / Video / Video segmentation and tracking
Segmenting, Fast and Slow: Real-Time Open-Vocabulary Video Instance Segmentation with Dual-Path Processing
Luca Barsellotti, Martin Sundermeyer, Mattia Segu, Nikita Araslanov, Muhammad Ferjad Naeem, Marcella Cornia, Yongqin Xian, Maxim Berman
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 Object Injection into P5 features (ablation) | SegFS fast path (MOBIUS MNv4-CM) · YouTubeVIS19 | AP: 37.9 -> 39.6 | Improved |
| add Object Injection into P5 features (ablation) | SegFS fast path (MOBIUS MNv4-CM) · OVIS | AP: 10.7 -> 12.5 | Improved |
| add learnable background token (ablation) | SegFS fast path · YouTubeVIS19 | AP: 39.6 -> 39.6 | No clear effect |
| add attention-based object projection (ablation) | SegFS fast path · YouTubeVIS19 | AP: 39.6 -> 40.2 | Improved |
| add attention-based object projection (ablation) | SegFS fast path · OVIS | AP: 12.8 -> 13.6 | Improved |
| add smoothing module (DSConvGN) after injection (ablation) | SegFS fast path · YouTubeVIS19 | AP: 40.2 -> 40.9 | Improved |
| add smoothing module (DSConvGN) after injection (ablation) | SegFS fast path · OVIS | AP: 13.6 -> 13.7 | No clear effect |
| increase top-K object embeddings beyond 50 (ablation) | SegFS · YouTubeVIS19 and OVIS | AP: peak at K=50, slightly lower plateau beyond | Hurt |
| increase propagation interval T from 1 to 10 (ablation) | SegFS, Reuse Objects, MPVSS vs Copy, RAFT, LiteFlowNet2 · all four datasets | AP: performance universally declines as T increases | Hurt |
| replace Reuse Objects with SegFS fast path | MOBIUS MNv4-CM · YouTubeVIS19 / FPS | AP / FPS: AP 42.1 -> 41.6; FPS 12.7 -> 38.2 | Mixed |
| replace Reuse Objects with SegFS fast path | MOBIUS ResNet50 · YouTubeVIS19 | AP: 50.3 -> 45.0 | Hurt |
| replace Reuse Objects with SegFS fast path | GLEE ResNet50 · YouTubeVIS19 | AP: 47.2 -> 41.8 | Hurt |
Closest papers by meaning
- Towards Open-Vocabulary Video Instance Segmentation (ICCV 2023 · similarity 0.92)
- OV-VIS: Open-Vocabulary Video Instance Segmentation (IJCV 2024 · similarity 0.90)
- OpenVIS: Open-vocabulary Video Instance Segmentation (AAAI 2025 · similarity 0.89)
- MobileInst: Video Instance Segmentation on the Mobile (AAAI 2024 · similarity 0.88)
- OW-VISCapTor: Abstractors for Open-World Video Instance Segmentation and Captioning (NeurIPS 2024 · similarity 0.88)
- Aligning Instance Brownian Bridge with Texts for Open-Vocabulary Video Instance Segmentation (AAAI 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.
