Research map / Video / Video segmentation and tracking
Test-Time Training on Video Streams
Renhao Wang, Yu Sun, Arnuv Tandon, Yossi Gandelsman, Xinlei Chen, Alexei A. Efros, Xiaolong Wang
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 implicit memory (init from previous model) to TTT-MAE (ablation) | TTT-MAE · COCO Videos instance segmentation | AP: 17.2 -> 18.3 | Improved |
| add explicit memory (sliding window) to TTT-MAE (ablation) | TTT-MAE · COCO Videos instance segmentation | AP: 17.2 -> 17.7 | Improved |
| add explicit memory on top of implicit memory (ablation) | Online TTT-MAE · COCO Videos panoptic segmentation | PQ: 22.7 -> 24.3 | Improved |
| add implicit memory on top of explicit memory (ablation) | Online TTT-MAE · KITTI-STEP test | mean IoU (%): 52.5 -> 54.5 | Improved |
| add explicit memory to implicit-only TTT (ablation) | Online TTT-MAE · KITTI-STEP val | mean IoU (%): 54.3 -> 55.6 | Improved |
| online TTT-MAE instead of Main Task Only | Mask2Former · COCO Videos instance segmentation | AP: 13.4 -> 19.4 | Improved |
| online TTT-MAE instead of Main Task Only | Mask2Former · COCO Videos panoptic segmentation | PQ: 14.6 -> 24.3 | Improved |
| online TTT-MAE instead of Main Task Only | Mask2Former · KITTI-STEP val | mean IoU (%): 53.8 -> 55.6 | Improved |
| online streaming TTT-MAE vs offline MAE on all frames | TTT-MAE · COCO Videos panoptic segmentation | PQ: 20.3 -> 24.3 | Improved |
| adapt normalization layers (LN Adapt, Tent) | Main Task Only · KITTI-STEP test | mean IoU (%): 52.5 -> 52.2 | No clear effect |
| joint training with MAE task | Main Task Only · COCO Videos instance segmentation | AP: 13.4 -> 13.0 | No clear effect |
| add temporal smoothing (ablation) | Main Task Only · KITTI-STEP val | mean IoU (%): +0.4% mean IoU | No clear effect |
Closest papers by meaning
- Test-Time Training Done Right (ICLR 2026 · similarity 0.91)
- Test Time Training for Long Videos via Frame Forgetting Network (ECCV 2026 · similarity 0.91)
- TTVidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining (NeurIPS 2026 · similarity 0.86)
- Learning What to Remember: Test-Time Training via Context Distillation (NeurIPS 2026 · similarity 0.85)
- In-Place Test-Time Training (ICLR 2026 · 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.
