Research map / Video / Video segmentation and tracking
Unidentified Video Objects: A Benchmark for Dense, Open-World Segmentation
Weiyao Wang, Matt Feiszli, Heng Wang, Du Tran
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 |
|---|---|---|---|
| finetune COCO-trained Mask R-CNN on UVO | Mask R-CNN · UVO frame-level | AR 100: 29.7 -> 41.3 | Improved |
| finetune COCO-trained Mask R-CNN on UVO | Mask R-CNN · COCO | AR 100: 49.6 -> 46.2 | Hurt |
| finetune LVIS-trained Mask R-CNN on UVO | Mask R-CNN · UVO frame-level | AR 100: 28.3 -> 39.0 | Improved |
| finetune YTVIS-trained MaskTrack R-CNN on UVO | MaskTrack R-CNN · UVO video-level | AR 100: 7.0 -> 13.2 | Improved |
| finetune on UVO, evaluate on YTVIS | MaskTrack R-CNN · YTVIS video-level | AR 100: 41.0 -> 42.3 | Improved |
| train with UVO only instead of YTVIS+UVO | MaskTrack R-CNN · UVO video-level | AR 100: 13.2 -> 17.2 | Improved |
| pretrain on ImageNet instead of Kinetics (ablation) | Mask R-CNN · UVO frame-level | AR 100: 37.3 -> 41.3 | Improved |
| pretrain on ImageNet instead of Kinetics (ablation) | MaskTrack R-CNN · UVO video-level | AR 100: 14.0 -> 23.0 | Improved |
| train only on a subset of categories (VOC) instead of all objects (ablation) | Mask R-CNN · Open-world UVO | Unseen AR: 8.8 (-39.4) | Hurt |
| train only on a subset of categories (ablation) | Mask R-CNN · Open-world UVO | Seen AR: 51.3 (+0.3) | No clear effect |
| train at 30fps instead of 6fps annotation, evaluate at 30fps (ablation) | MaskTrack R-CNN · UVO video 30fps test | AR: 14.5 -> 17.2 | Improved |
| use 1fps annotation plus interpolation instead of 6fps training (ablation) | MaskTrack R-CNN · UVO video 30fps test | AR: 14.5 -> 15.0 | Mixed |
Closest papers by meaning
- Video Instance Segmentation in an Open-World (IJCV 2024 · similarity 0.90)
- Learning Unsupervised Video Object Segmentation Through Visual Attention (CVPR 2019 · similarity 0.89)
- Reciprocal Transformations for Unsupervised Video Object Segmentation (CVPR 2021 · similarity 0.89)
- Breaking the "Object" in Video Object Segmentation (CVPR 2023 · similarity 0.88)
- OVT-B: A New Large-Scale Benchmark for Open-Vocabulary Multi-Object Tracking (NeurIPS 2024 · similarity 0.87)
- GMOS: Grounding Moving Object Segmentation in 3D Space and Time (NeurIPS 2026 · similarity 0.87)
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.
