Research map / Video / Video segmentation and tracking
FSD-Net: Foundation-Guided Spatiotemporal Distillation for Video Polyp Segmentation
Yahang Leng, Shiquan Min, Chengzhou Li
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 semantic flow distillation instead of optical flow propagation (ablation) | FSD-Net baseline · SUN-SEG-Hard | mIoU: 64.88 -> 70.15 | Improved |
| add plain spatial knowledge distillation instead of optical flow (ablation) | FSD-Net baseline · SUN-SEG-Hard | mIoU: 64.88 -> 67.52 | Improved |
| add standard cycle consistency on top of SFD (ablation) | FSD-Net · SUN-SEG-Hard | mDice: 79.42 -> 78.56 | Hurt |
| add occlusion-aware cycle consistency (OACC) on top of SFD (ablation) | FSD-Net · SUN-SEG-Hard | mDice: 79.42 -> 85.31 | Improved |
| add static prototype re-identification (ablation) | FSD-Net (SFD+OACC) · SUN-SEG-Hard | mDice: 85.31 -> 85.35 | No clear effect |
| add momentum-updated PRM instead of static prototype (ablation) | FSD-Net (SFD+OACC) · SUN-SEG-Hard | mDice: 85.31 -> 87.04 | Improved |
| use all three modules SFD, OACC and PRM together instead of a single module (ablation) | FSD-Net · SUN-SEG-Hard | mDice: 79.56 -> 87.04 | Improved |
| prompt only the first frame instead of the middle frame only (ablation) | FSD-Net · SUN-SEG-Hard | mDice: 83.20 -> 87.04 | Improved |
| add more prompted frames (first, middle, last) beyond the first frame (ablation) | FSD-Net · SUN-SEG-Hard | mDice: 87.04 -> 86.10 | Hurt |
| supply one point prompt per frame at test time instead of prompt-free inference | FSD-Net · SUN-SEG-Hard | mDice: 85.28 -> 87.04 | Improved |
| set visibility threshold to 0.80 instead of 0.60 (ablation) | OACC · SUN-SEG-Hard | mDice: 84.66 -> 87.04 | Improved |
| FSD-Net with sparse annotation versus fully supervised Polyp-SAM-FS | FSD-Net · SUN-SEG-Hard | mDice: 80.52 -> 87.04 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VP-SAM — VP-SAM: Taming Segment Anything Model for Video Polyp Segmentation via Disentanglement and Spatio-temporal Side Network (ECCV 2024) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- FST-SAM3: Taming SAM~3 with Frequency-Spatio-Temporal Refinement for Video Polyp Segmentation (ECCV 2026 · similarity 0.90)
- HFSTI-Net: Hierarchical Frequency-spatial-temporal Interactions for Video Polyp Segmentation (ICLR 2026 · similarity 0.89)
- An Embedding-Unleashing Video Polyp Segmentation Framework via Region Linking and Scale Alignment (AAAI 2024 · similarity 0.86)
- Boost the Inference with Co-training: A Depth-guided Mutual Learning Framework for Semi-supervised Medical Polyp Segmentation (CVPR 2025 · similarity 0.85)
- Polyper: Boundary Sensitive Polyp Segmentation (AAAI 2024 · similarity 0.84)
- Weakly Supervised Gland Segmentation with Class Semantic Consistency and Purified Labels Filtration (AAAI 2025 · similarity 0.83)
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.
