Research map / Video / Video segmentation and tracking
SegFlow: Joint Learning for Video Object Segmentation and Optical Flow
Jingchun Cheng, Yi-Hsuan Tsai, Shengjin Wang, Ming-Hsuan Yang
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 flow branch with bi-directional feature propagation (ablation) | SegFlow segmentation branch · DAVIS val (semi-supervised) | J mean: 0.724 -> 0.748 | Improved |
| add the flow branch with bi-directional feature propagation (ablation) | SegFlow segmentation branch · DAVIS val (semi-supervised) | T mean (lower is better): 0.250 -> 0.194 | Improved |
| add the flow branch (offline only) (ablation) | SegFlow segmentation branch · DAVIS val (unsupervised) | J mean: 0.654 -> 0.674 | Improved |
| add offline training before online fine-tuning (ablation) | SegFlow · DAVIS val (SegFlow ablation) | J mean: 0.538 -> 0.748 | Improved |
| add online fine-tuning with first-frame mask and iterative training (ablation) | SegFlow · DAVIS val (SegFlow ablation) | J mean: 0.669 -> 0.748 | Improved |
| add flow data augmentation by synthesizing the next frame (ablation) | SegFlow · DAVIS val (SegFlow ablation) | T mean (lower is better): 0.225 -> 0.194 | Improved |
| add flow data augmentation (ablation) | SegFlow · DAVIS val (SegFlow ablation) | J mean: 0.739 -> 0.748 | Improved |
| add segmentation data augmentation (ablation) | SegFlow without flow, offline · DAVIS val (SegFlow ablation) | J mean: 0.606 -> 0.654 | Improved |
| average predictions of image and its flipped version | SegFlow · DAVIS val | J mean: 0.748 -> 0.761 | Improved |
| replace FlowNetS+ft* with SegFlow (add segmentation branch) for flow | FlowNetS · Sintel Final train | average endpoint error: 4.26 -> 2.61 | Improved |
| replace FlowNetS+ft* with SegFlow for flow | FlowNetS · KITTI train | average endpoint error: 7.37 -> 4.40 | Improved |
| replace FlowNetS+ft (published) with SegFlow for flow | FlowNetS · Sintel Clean test | average endpoint error: 6.16 -> 7.45 | Hurt |
Closest papers by meaning
- Video Segmentation via Object Flow (CVPR 2016 · similarity 0.89)
- Every Frame Counts: Joint Learning of Video Segmentation and Optical Flow (AAAI 2020 · similarity 0.87)
- Learning to Segment Moving Objects (IJCV 2018 · similarity 0.87)
- UnSAMFlow: Unsupervised Optical Flow Guided by Segment Anything Model (CVPR 2024 · similarity 0.86)
- FlowFeat: Pixel-Dense Embedding of Motion Profiles (NeurIPS 2025 · similarity 0.86)
- Learning Segmentation from Point Trajectories (NeurIPS 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.
