Research map / Detection and segmentation / Open-set and few-shot detection
Object Discovery in Videos as Foreground Motion Clustering
Christopher Xie, Yu Xiang, Zaid Harchaoui, Dieter Fox
Check 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 learned pixel-trajectory clustering rather than clustering foreground pixels independently in each frame (ablation) | Object Discovery in Videos as Foreground Motion Clustering · FBMS | multi-object F: 59.7 -> 67.3 | Improved |
| Use foreground masks to restrict trajectory clustering (ablation) | Object Discovery in Videos as Foreground Motion Clustering · FBMS | multi-object F: 59.6 -> 67.3 | Improved |
| Add the spatial coordinate module to trajectory embeddings (ablation) | Object Discovery in Videos as Foreground Motion Clustering · FBMS | multi-object F: 63.2 -> 67.3 | Improved |
| Pre-train on FT3D before FBMS and DAVIS-m training (ablation) | Object Discovery in Videos as Foreground Motion Clustering · FBMS | multi-object F: 63.1 -> 67.3 | Improved |
| Add DAVIS-m to FBMS training (ablation) | Object Discovery in Videos as Foreground Motion Clustering · FBMS | multi-object F: 62.1 -> 67.3 | Improved |
| Use mid-level RGB/optical-flow fusion rather than early fusion (ablation) | Y-Net · FT3D | foreground IoU: 0.883 -> 0.905 | Improved |
| Use mid-level RGB/optical-flow fusion rather than late fusion (ablation) | Y-Net · FT3D | foreground IoU: 0.897 -> 0.905 | Improved |
| Use conv PT-RNN instead of convGRU PT-RNN (ablation) | Pixel Trajectory RNN · FBMS | multi-object F: 64.8 -> 67.3 | Improved |
| Use conv PT-RNN instead of standard PT-RNN (ablation) | Pixel Trajectory RNN · FBMS | multi-object F: 66.0 -> 67.3 | Improved |
| Use the proposed method rather than CCG | Multi-object motion segmentation · FBMS | multi-object F: 65.0 -> 67.3 | Improved |
| Use the proposed method rather than CCG | Multi-object motion segmentation · ComplexBackground | multi-object F: 65.6 -> 58.3 | Hurt |
| Use the proposed method rather than CCG | Multi-object motion segmentation · CamouflagedAnimal | multi-object F: 72.2 -> 75.3 | Improved |
Closest papers by meaning
- Discovering Objects That Can Move (CVPR 2022 · similarity 0.89)
- Learning Motion Patterns in Videos (CVPR 2017 · similarity 0.89)
- Learning Video Object Segmentation With Visual Memory (ICCV 2017 · similarity 0.89)
- Learning Segmentation from Point Trajectories (NeurIPS 2024 · similarity 0.89)
- Super-Trajectory for Video Segmentation (ICCV 2017 · similarity 0.89)
- Segment Any Motion in Videos (CVPR 2025 · 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.
