Research map / Video / Video segmentation and tracking
Segmenting the Motion Components of a Video: A Long-Term Unsupervised Model
Etienne Meunier, Patrick Bouthémy
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 the transformer decoder on the 3D U-Net latent space (ablation) | LT-MS-K4 · DAVIS2016 | Jaccard J: 71.3 -> 72.4 | Improved |
| add the transformer decoder on the 3D U-Net latent space (ablation) | LT-MS-K4 · FBMS59 | Jaccard J: 55.5 -> 58.2 | Improved |
| add the temporal consistency loss term Lc (ablation) | LT-MS-K4 · SegTrackV2 | Jaccard J: 57.5 -> 60.4 | Improved |
| add the temporal consistency loss term Lc (ablation) | LT-MS-K4 · DAVIS2016 | Jaccard J: 71.0 -> 72.4 | Improved |
| use B-spline temporal motion model instead of polynomial space-time model (ablation) | LT-MS-K4 · DAVIS2016 | Jaccard J: 69.8 -> 72.4 | Improved |
| use B-spline temporal motion model instead of polynomial space-time model (ablation) | LT-MS-K4 · FBMS59 | Jaccard J: 54.5 -> 58.2 | Improved |
| use the L1-based consistency loss instead of the pre-approximation loss of eq. (11) (ablation) | LT-MS-K4 · SegTrackV2 | Jaccard J: 57.4 -> 60.4 | Improved |
| use the L1-based consistency loss instead of the pre-approximation loss of eq. (11) (ablation) | LT-MS-K4 · DAVIS2016 | Jaccard J: 71.6 -> 72.4 | No clear effect |
| train on real DAVIS2016 data instead of synthetic FT3D | LT-MS-K4 · DAVIS2016, FBMS59, SegTrackV2 | Jaccard J: 72.4 -> 72.8 | No clear effect |
| use long-term LT-MS instead of frame-based EM, sequence-based evaluation | LT-MS-K4 · DAVIS2016 | Jaccard J: 68.0 -> 72.4 | Improved |
| use long-term LT-MS instead of frame-based EM, frame-based evaluation | LT-MS-K4 · DAVIS2016 | Jaccard J: 77.1 -> 75.0 | Hurt |
| lengthen the input flow sequence at inference from 10 | LT-MS-K4 · DAVIS2016, FBMS59, SegTrackV2 | Jaccard J | No clear effect |
Closest papers by meaning
- Unsupervised Space-Time Network for Temporally-Consistent Segmentation of Multiple Motions (CVPR 2023 · similarity 0.94)
- Learning Segmentation from Point Trajectories (NeurIPS 2024 · similarity 0.91)
- Learning Motion Patterns in Videos (CVPR 2017 · similarity 0.90)
- Video Segmentation via Object Flow (CVPR 2016 · similarity 0.90)
- Learning to Segment Moving Objects (IJCV 2018 · similarity 0.89)
- 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.
