Research map / Video / Video segmentation and tracking
Deep Transport Network for Unsupervised Video Object Segmentation
Kaihua Zhang, Zicheng Zhao, Dong Liu, Qingshan Liu, Bo Liu
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 |
|---|---|---|---|
| replace dense Sinkhorn with LSH only (ablation) | TransportNet · DAVIS-16 / FBMS | mean J: 83.4 -> 82.4 (DAVIS-16), 75.0 -> 73.2 (FBMS) | Hurt |
| replace dense Sinkhorn with SSH only (ablation) | TransportNet · DAVIS-16 / FBMS | mean J: 83.4 -> 83.9 (DAVIS-16), 75.0 -> 77.4 (FBMS) | Improved |
| use full LSSH (LSH then SSH) instead of dense Sinkhorn (ablation) | TransportNet · DAVIS-16 / FBMS | mean J: 83.4 -> 84.5 (DAVIS-16), 75.0 -> 78.7 (FBMS) | Improved |
| add one LSSH after Res5 (ablation) | TransportNet · DAVIS-16 / FBMS | mean J: 80.4 -> 83.3 (DAVIS-16), 68.2 -> 72.7 (FBMS) | Improved |
| add a second LSSH after Res4 (ablation) | TransportNet · DAVIS-16 / FBMS | mean J: 83.3 -> 83.8 (DAVIS-16), 72.7 -> 76.4 (FBMS) | Improved |
| add a third LSSH after Res3 (ablation) | TransportNet · DAVIS-16 / FBMS | mean J: 83.8 -> 84.5 (DAVIS-16), 76.4 -> 78.7 (FBMS) | Improved |
| add edge loss (ablation) | TransportNet · DAVIS-16 / FBMS | mean J: 84.3 -> 84.5 (DAVIS-16), 76.3 -> 78.7 (FBMS) | Improved |
| use LSSH instead of dense Sinkhorn for feature matching (ablation) | TransportNet · 128x128x2048 feature map | GPU memory and time: 15.8% GPU memory, 24.5x faster | Improved |
| use TransportNet instead of 3DC-Seg | UVOS on FBMS · FBMS test | mean J: 76.1 -> 78.7 (vs MATNet) | Improved |
Closest papers by meaning
- Learning to Segment Moving Objects (IJCV 2018 · similarity 0.90)
- Learning Video Object Segmentation With Visual Memory (ICCV 2017 · similarity 0.89)
- Unsupervised Video Object Segmentation with Motion-based Bilateral Networks (ECCV 2018 · similarity 0.89)
- Motion-Guided Cascaded Refinement Network for Video Object Segmentation (CVPR 2018 · similarity 0.88)
- Anchor Diffusion for Unsupervised Video Object Segmentation (ICCV 2019 · similarity 0.88)
- Temporally Consistent Unbalanced Optimal Transport for Unsupervised Action Segmentation (CVPR 2024 · 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.
