Research map / Detection and segmentation / Open-set and few-shot detection
Discovering Objects That Can Move
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 sparse motion segmentation supervision (ablation) | Ours · TRI-PD validation | Fg. ARI All: 13.1 -> 50.9 | Helped |
| Use ground-truth segments for moving and static objects instead of estimated motion segments (ablation) | Ours · TRI-PD validation | Fg. ARI All: 50.9 -> 71.7 | Helped |
| Use only ground-truth moving-object masks as supervision (ablation) | Ours · TRI-PD validation | Fg. ARI All: 13.1 -> 59.6 | Helped |
| Use learning-based motion segmentation [14] rather than heuristic segmentation [33] on ground-truth flow (ablation) | Ours · TRI-PD validation | Fg. ARI All: 42.8 -> 51.7 | Helped |
| Replace ground-truth flow with RAFT-estimated flow for [14] (ablation) | Ours · TRI-PD validation | Fg. ARI All: 51.7 -> 50.9 | No clear effect |
| Replace RAFT flow with SMURF-estimated flow for [14] (ablation) | Ours · TRI-PD validation | Fg. ARI All: 50.9 -> 50.5 | No clear effect |
| Add ConvGRU video memory and train on video clips rather than individual frames (ablation) | SlotAttention-based model · CATER validation | Fg. ARI: 66.3 -> 71.5 | Helped |
| Replace iterative inference with one-shot attention and learnable initialization (ablation) | Video object-discovery model · CATER validation | Fg. ARI: 71.5 -> 83.2 | Helped |
| Add temporal consistency regularization (ablation) | Video object-discovery model · CATER validation | Fg. ARI: 83.2 -> 86.7 | Helped |
| Replace per-slot decoding with one-shot decoding without motion supervision (ablation) | Video object-discovery model · CATER validation | Fg. ARI: 86.7 -> 34.5 | Hurt |
| Add motion supervision to one-shot decoding (ablation) | Video object-discovery model · CATER validation | Fg. ARI: 34.5 -> 92.7 | Helped |
| Add the reconstruction objective to the motion-supervised model (ablation) | Video object-discovery model · CATER validation | Fg. ARI: 77.9 -> 92.7 | Helped |
Closest papers by meaning
- Object Concepts Emerge from Motion (NeurIPS 2025 · similarity 0.88)
- GMOS: Grounding Moving Object Segmentation in 3D Space and Time (NeurIPS 2026 · similarity 0.88)
- Segment Any Motion in Videos (CVPR 2025 · similarity 0.87)
- Learning Segmentation from Point Trajectories (NeurIPS 2024 · similarity 0.87)
- Cross-Modal Distillation for 2D/3D Multi-Object Discovery from 2D Motion (CVPR 2025 · similarity 0.87)
- DIOD: Self-Distillation Meets Object Discovery (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.