Research map / Video / Video segmentation and tracking
Test Time Training for Long Videos via Frame Forgetting Network
Rajat modi, Xin Liang, Sebastian Noel, Yogesh Rawat
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 sliding-window TTT-MAE with FFN (3 frames per step) | Online TTT-MAE · COCO-Videos | instance segmentation AP: 37.6 -> 45.1 | Improved |
| replace sliding-window TTT-MAE with FFN | Online TTT-MAE · COCO-Videos | time per frame (s): 4.1 -> 0.7 | Improved |
| replace sliding-window TTT-MAE with FFN | Online TTT-MAE · KITTI-STEP test | mIoU: 54.3 -> 59.5 | Improved |
| replace sliding-window TTT-Online with FFN on hours-long videos | TTT-Online (s.w.) · EpicTours | semantic segmentation mIoU: 42.8 -> 49.3 | Improved |
| replace Online TTT with FFN for action classification | Online TTT on DINOv3 · UCF101 | top-1 accuracy: 94.2 -> 95.0 | Improved |
| replace Online TTT with FFN for action classification | Online TTT on DINOv3 · SSv2 | top-1 accuracy: 73.4 -> 74.1 | Improved |
| add timestep conditioning of backbone (ablation) | FFN · COCO-Videos | instance segmentation AP: 42.7 -> 45.1 | Improved |
| use RoPE instead of absolute time encoding (ablation) | FFN · COCO-Videos | instance segmentation AP: 45.1 -> 46.8 | Improved |
| use absolute instead of relative time encoding (ablation) | FFN · COCO-Videos | instance segmentation AP: 40.5 -> 45.1 | Improved |
| use MBO (merge) buffer instead of FIFO (ablation) | FFN · COCO-Videos | instance segmentation AP: 43.9 -> 45.1 | Improved |
| combine FIFO and MBO buffer policies (ablation) | FFN · COCO-Videos | instance segmentation AP: 45.1 -> 45.7 | Improved |
| use cosine instead of L1 for latent similarity (ablation) | FFN · COCO-Videos | instance segmentation AP: 43.2 -> 45.1 | Improved |
Closest papers by meaning
- Learning What to Remember: Test-Time Training via Context Distillation (NeurIPS 2026 · similarity 0.87)
- TTVidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining (NeurIPS 2026 · similarity 0.86)
- Unleashing Hour-Scale Video Training for Long Video-Language Understanding (NeurIPS 2025 · similarity 0.86)
- ReWind: Understanding Long Videos with Instructed Learnable Memory (CVPR 2025 · similarity 0.86)
- Re-thinking Temporal Search for Long-Form Video Understanding (CVPR 2025 · similarity 0.85)
- One-Minute Video Generation with Test-Time Training (CVPR 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.
