Research map / Video / Video segmentation and tracking
Made to Order: Discovering monotonic temporal changes via self-supervised video ordering
Charig Yang, Weidi Xie, Andrew Zisserman
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 |
|---|---|---|---|
| use reversible (min of forward and reverse) ordering loss (ablation) | Made to Order · MoCA (val) | exact match: 44.5 -> 82.0 | Improved |
| use divided space-time attention instead of space-only attention in the encoder (ablation) | Made to Order · MUDS (4 frames) | exact match: 13.0 -> 56.4 | Improved |
| use divided space-time attention instead of full attention (ablation) | Made to Order · MUDS (4 frames) | exact match: 7.6 -> 56.4 | Improved |
| use space-only attention instead of divided space-time on image-set ordering (ablation) | Made to Order · SVHN (9 frames) | exact match: 53.9 -> 55.8 | No clear effect |
| double encoder size (ablation) | Made to Order · SVHN (9 frames) | exact match: 53.9 -> 62.9 | Improved |
| double decoder size (ablation) | Made to Order · SVHN (9 frames) | exact match: 53.9 -> 53.9 | No clear effect |
| reduce patch size from 6 to 3 (ablation) | Made to Order · SVHN (9 frames) | exact match: 53.9 -> 47.5 | Hurt |
| replace patch projection with pretrained DINOv2 | Made to Order · Monotonic MUDS | segmentation mIoU: 37.9 -> 41.3 | Improved |
| prompt SAM with attribution map peaks | Made to Order · Monotonic MUDS | segmentation mIoU: 37.9 -> 45.1 | Improved |
| use ordering proxy task instead of arrow-of-time proxy (same architecture) (ablation) | Made to Order · Monotonic MUDS | localization accuracy: 63.3 -> 83.3 | Improved |
| use 3 or 4 frames instead of 5 frames per sequence (ablation) | Made to Order · Timelapse clocks (cropped) | exact match: 2 frames 51.3, 3 frames 65.2, 4 frames 62.5, 5 frames 37.1 | Mixed |
| train and test on same clips versus separate videos (ablation) | Made to Order · MUDS | exact match: 45.1 -> 88.5 | Improved |
Closest papers by meaning
- Graph Representation for Order-Aware Visual Transformation (CVPR 2023 · similarity 0.87)
- REOrdering Patches Improves Vision Models (NeurIPS 2025 · similarity 0.86)
- Zero-Shot Scene Change Detection (AAAI 2025 · similarity 0.85)
- Learning Object State Changes in Videos: An Open-World Perspective (CVPR 2024 · similarity 0.84)
- Attention (as Discrete-Time Markov) Chains (NeurIPS 2025 · similarity 0.83)
- Change3D: Revisiting Change Detection and Captioning from A Video Modeling Perspective (CVPR 2025 · similarity 0.83)
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.
