Research map / Video / Video segmentation and tracking
Seeing as Humans Do: Learning from Motion to Segment Anything Without Supervision
Weijian Jian, Xiaoyue Zhang, Bin Xiao, Chunyu Xie, Yixiao He, Yutao Liu, Dawei Leng, Yuhui Yin
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 PGCL loss instead of BCE+Dice (ablation) | PGM · SA-1B 1000 images | AR (%): 3.8 -> 28.3 | Improved |
| add PGCL loss instead of Slot Attention (ablation) | PGM · SA-1B 1000 images | AR (%): 8.2 -> 28.3 | Improved |
| add MFSM to ROSM (ablation) | MGMS pipeline · SA-1B 1000 images | AR (%): 28.8 -> 36.6 | Improved |
| add ROSM to MFSM (ablation) | MGMS pipeline · SA-1B 1000 images | AR (%): 34.7 -> 36.6 | Improved |
| combine maskness and boundary sharpness scores (ablation) | MGMS mask filtering · Whole-image Avg | AR (%): 24.1 -> 28.8 | Improved |
| combine maskness with sharpness versus sharpness alone (ablation) | MGMS mask filtering · Whole-image Avg | AR (%): 22.8 -> 28.8 | Improved |
| increase prediction heads from one to four (ablation) | PGM · SA-1B 1000 images | AR (%): 27.9 -> 36.6 | Improved |
| increase prediction heads from four to five (ablation) | PGM · SA-1B 1000 images | AR (%): 36.6 -> 36.5 | No clear effect |
| increase video pre-training data from 100h to 1,000h (ablation) | PGM · SA-1B 1000 images | AR (%): 16.8 -> 28.3 | Improved |
| increase video pre-training data from 1,000h to 10,000h (ablation) | PGM · SA-1B 1000 images | AR (%): 28.3 -> 36.6 | Improved |
| add CRF refinement to PGM pseudo-masks | PGM · Whole-image Avg | AR (%): 28.8 -> 30.1 | Improved |
| use PGM unrefined masks instead of UnSAM CascadePSP pseudo-labels | PGM · Whole-image Avg | AR (%): 21.7 -> 28.8 | Improved |
Closest papers by meaning
- Segment Anything without Supervision (NeurIPS 2024 · similarity 0.89)
- GMOS: Grounding Moving Object Segmentation in 3D Space and Time (NeurIPS 2026 · similarity 0.88)
- Object Concepts Emerge from Motion (NeurIPS 2025 · similarity 0.87)
- Perceive Anything: Recognize, Explain, Caption, and Segment Anything in Images and Videos (NeurIPS 2025 · similarity 0.87)
- Segment Any Motion in Videos (CVPR 2025 · similarity 0.87)
- Matching Anything by Segmenting Anything (CVPR 2024 · 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.
