Research map / Video / Video segmentation and tracking
Live Interactive Training for Video Segmentation
Xinyu Yang, Haozheng Yu, Yihong Sun, Bharath Hariharan, Jennifer J. Sun
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 LIT-LoRA online adaptation | SAM2 · VOST | corrections per video: 27.43 -> 18.24 | Improved |
| add LIT-LoRA online adaptation | SAM2 · VOST | annotation time (min): 18.42 -> 12.91 | Improved |
| add LIT-LoRA at stricter threshold tau_IoU=0.75 | SAM2 · datasets | corrections: 17-24% reduction | Improved |
| add LIT to DAM4SAM | DAM4SAM · VOST | corrections: 34.60 -> 22.46 | Improved |
| add LIT to SAMURAI | SAMURAI · VOST | corrections: 26.96 -> 21.23 | Improved |
| add LIT-LoRA to CLIP online classification | CLIP ViT-B/32 · Stanford Cars | corrections per class: 13.38 -> 7.57 | Improved |
| fine-tune original decoder at each correction instead of separate LoRA (ablation) | LIT · VOST | corrections per video: 27.43 -> 32.47 | Hurt |
| continual LoRA learning across corrections (vs first correction only) (ablation) | LIT-LoRA · VOST | corrections per video: 21.43 -> 18.24 | Improved |
| full decoder fine-tuning instead of LoRA (ablation) | LIT · VOST | corrections per video: 18.24 -> 17.46 | Improved |
| maintain three LoRAs instead of one (ablation) | LIT-LoRA · VOST | corrections per video: 18.24 -> 17.87 | No clear effect |
| increase training epochs from 40 to 60-100 (ablation) | LIT-LoRA · VOST | correction reduction: < 2 pp gain | No clear effect |
| add LIT-LoRA in user study with 6 volunteers | SAM2 demo · VOST user study | correction number: 41.92% fewer corrections | Improved |
Closest papers by meaning
- Continuous Adaptation for Interactive Object Segmentation by Learning from Corrections (ECCV 2020 · similarity 0.88)
- Convolution Meets LoRA: Parameter Efficient Finetuning for Segment Anything Model (ICLR 2024 · similarity 0.86)
- C-LoRA: Continual Low-Rank Adaptation for Pre-trained Visual Models (NeurIPS 2026 · similarity 0.86)
- Rethinking Interactive Image Segmentation with Low Latency High Quality and Diverse Prompts (CVPR 2024 · similarity 0.85)
- SAM-I2V: Upgrading SAM to Support Promptable Video Segmentation with Less than 0.2% Training Cost (CVPR 2025 · similarity 0.85)
- SAM 2: Segment Anything in Images and Videos (ICLR 2025 · similarity 0.85)
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.
