Research map / Video / Video segmentation and tracking
Hierarchical Feature Alignment Network for Unsupervised Video Object Segmentation
Gensheng Pei, Fumin Shen, Yazhou Yao, Guo-Sen Xie, Zhenmin Tang, Jinhui Tang
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 optical flow to the image frame input (baseline fusion vs image only) (ablation) | baseline additive fusion · DAVIS-16 val (MiT-b1) | J Mean: 79.1 -> 83.0 | Improved |
| add the image frame to optical flow input (baseline fusion vs flow only) (ablation) | baseline additive fusion · DAVIS-16 val (MiT-b1) | J Mean: 77.9 -> 83.0 | Improved |
| add the feature alignment module (FAM) (ablation) | baseline · DAVIS-16 val (MiT-b1) | J Mean: 83.0 -> 85.2 | Improved |
| add the feature adaptation module (FAT) (ablation) | baseline · DAVIS-16 val (MiT-b1) | J Mean: 83.0 -> 85.0 | Improved |
| add both FAM and FAT (ablation) | baseline · DAVIS-16 val (MiT-b1) | J Mean: 83.0 -> 86.2 | Improved |
| add FAM, FAT and both modules costs speed (ablation) | baseline · DAVIS-16 val (MiT-b1) | FPS: 22.0 -> 20.8 | Hurt |
| enlarge backbone from MiT-b0 to MiT-b1 (ablation) | HFAN · DAVIS-16 val (MiT-b1) single-scale | J Mean: 81.5 -> 86.2 | Improved |
| enlarge backbone from MiT-b1 to MiT-b3 (ablation) | HFAN · DAVIS-16 val single-scale | FPS: 20.8 -> 10.6 | Hurt |
| use multi-scale test instead of single-scale test (ablation) | HFAN MiT-b1 · DAVIS-16 val | J Mean: 86.2 -> 87.1 | Improved |
| raise input size from 384x384 to 512x512 (RAFT flow) (ablation) | HFAN MiT-b1 · DAVIS-16 val multi-scale | J Mean: 86.2 -> 87.1 | Improved |
| use RAFT instead of PWCNet optical flow (ablation) | HFAN MiT-b1 · DAVIS-16 val multi-scale 512x512 | J Mean: 85.7 -> 87.1 | Improved |
| use HFAN instead of MATNet+CRF with the same MiT-b1 backbone | MiT-b1 · DAVIS-16 val | J Mean: 83.8 -> 87.1 | Improved |
Closest papers by meaning
- Deep Transport Network for Unsupervised Video Object Segmentation (ICCV 2021 · similarity 0.88)
- Reciprocal Transformations for Unsupervised Video Object Segmentation (CVPR 2021 · similarity 0.87)
- Sfnet: Faster and Accurate Semantic Segmentation Via Semantic Flow (IJCV 2023 · similarity 0.86)
- Video Segmentation via Object Flow (CVPR 2016 · similarity 0.86)
- Video Object Segmentation With Dynamic Memory Networks and Adaptive Object Alignment (ICCV 2021 · similarity 0.86)
- Semantic and Sequential Alignment for Referring Video Object Segmentation (CVPR 2025 · similarity 0.84)
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.
