Research map / Video / Video segmentation and tracking
Learning To Better Segment Objects From Unseen Classes With Unlabeled Videos
Yuming Du, Yang Xiao, Vincent Lepetit
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 background loss L_I (ablation) | mask selection · Unseen-VIS-test | AP: 35.8 -> 36.6 | Improved |
| add non-overlapping constraint (ablation) | L_I selection · Unseen-VIS-test | AP: 36.6 -> 38.1 | Improved |
| add flow loss L_F (ablation) | L_I + constraint · Unseen-VIS-test | AP: 38.1 -> 38.7 | Improved |
| add regularization loss L_p (ablation) | L_I + constraint + L_F · Unseen-VIS-test | AP: 38.7 -> 39.0 | Improved |
| add background loss L_I (ablation) | mask selection · Unseen-VIS-test | AP75: 38.1 -> 37.8 | No clear effect |
| select masks with our method vs data distillation | MP R-CNN fine-tuning · Unseen-VIS-test | AP: 36.6 -> 39.0 | Improved |
| select masks with our method vs UnOVOST tracking | MP R-CNN fine-tuning · Unseen-VIS-test | AP: 31.1 -> 39.0 | Improved |
| select masks with our method vs RVOS (video annotated) | MP R-CNN fine-tuning · Unseen-VIS-test | AP75: 38.0 -> 41.3 | Improved |
| use ground-truth first-frame masks (OS-UVC) vs ours | mask generation · Unseen-VIS-test | AP: 39.0 vs 41.5 | Hurt |
| replace GT first-frame masks with predicted masks in UVC | UVC · Unseen-VIS-test | AP: 41.5 -> 21.2 | Hurt |
| fine-tune UnOVOST mask network with our masks | UnOVOST · DAVIS | J&F mean: 56.2 -> 59.9 | Improved |
| retrain with COCO plus our masks | MP R-CNN · Unseen-VIS-test | AP: 35.8 -> 38.9 | Improved |
Closest papers by meaning
- Unsupervised Object Segmentation in Video by Efficient Selection of Highly Probable Positive Features (ICCV 2017 · similarity 0.90)
- UniT: Unified Knowledge Transfer for Any-Shot Object Detection and Segmentation (CVPR 2021 · similarity 0.89)
- Self-supervised Amodal Video Object Segmentation (NeurIPS 2022 · similarity 0.88)
- SegPrompt: Boosting Open-World Segmentation via Category-Level Prompt Learning (ICCV 2023 · similarity 0.88)
- Unidentified Video Objects: A Benchmark for Dense, Open-World Segmentation (ICCV 2021 · similarity 0.88)
- Self-supervised Object-Centric Learning for Videos (NeurIPS 2023 · 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.
