Research map / Video / Video segmentation and tracking
See More, Know More: Unsupervised Video Object Segmentation With Co-Attention Siamese Networks
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 co-attention to the frame-wise feature embedding model (ablation) | COSNet · DAVIS16 | mean J: 71.3 -> 80.5 | Helped |
| use symmetric instead of vanilla co-attention (ablation) | COSNet · DAVIS16, FBMS, YouTube-Objects | mean J: 80.0 -> 80.5; 75.2 -> 75.6; 70.3 -> 70.5 | Helped |
| use symmetric instead of channel-wise co-attention (ablation) | COSNet · DAVIS16, FBMS, YouTube-Objects | mean J: 77.2 -> 80.5; 72.7 -> 75.6; 67.5 -> 70.5 | Helped |
| fuse reference information in attention-summary feature space rather than averaging predictions (ablation) | COSNet · DAVIS16, FBMS, YouTube-Objects | mean J: 79.5 -> 80.5; 74.2 -> 75.6; 69.9 -> 70.5 | Helped |
| use global uniform rather than local consecutive reference-frame sampling (ablation) | COSNet · DAVIS16, FBMS, YouTube-Objects | mean J: 80.26 -> 80.53; 75.52 -> 75.61; 70.43 -> 70.54 | Helped |
| use global random rather than global uniform reference-frame sampling (ablation) | COSNet · DAVIS16, FBMS, YouTube-Objects | mean J: 80.53 -> 80.52; 75.61 -> 75.54; 70.54 -> 70.55 | No clear effect |
| increase reference frames from 0 to 1 (ablation) | COSNet · DAVIS16 | mean J: 71.3 -> 77.6 | Helped |
| increase reference frames from 1 to 5 (ablation) | COSNet · DAVIS16 | mean J: 77.6 -> 80.5 | Helped |
| increase reference frames from 5 to 7 (ablation) | COSNet · DAVIS16 | mean J: 80.5 -> 80.5 | No clear effect |
| use COSNet rather than PDB | DAVIS16 · DAVIS16 test | mean J: 77.2 -> 80.5 | Helped |
| use COSNet rather than PDB | DAVIS16 · DAVIS16 test | mean F: 74.5 -> 79.4 | Helped |
| use COSNet rather than IET | FBMS · FBMS test | mean J: 71.9 -> 75.6 | Helped |
Closest papers by meaning
- DeepCO3: Deep Instance Co-Segmentation by Co-Peak Search and Co-Saliency Detection (CVPR 2019 · similarity 0.89)
- F2Net: Learning to Focus on the Foreground for Unsupervised Video Object Segmentation (AAAI 2021 · similarity 0.88)
- Deep Object Co-Segmentation via Spatial-Semantic Network Modulation (AAAI 2020 · similarity 0.88)
- Prototypical Cross-Attention Networks for Multiple Object Tracking and Segmentation (NeurIPS 2021 · similarity 0.88)
- Siamese Network with Interactive Transformer for Video Object Segmentation (AAAI 2022 · similarity 0.88)
- Dual Prototype Attention for Unsupervised Video Object Segmentation (CVPR 2024 · similarity 0.86)
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.