Research map / Detection and segmentation / CNN features
ViT$^3$: Unlocking Test-Time Training in Vision
Dongchen Han, Yining Li, Tianyu Li, Zixuan Cao, Ziming Wang, Jun Song, Yu Cheng, Bo Zheng, Gao Huang
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 |
|---|---|---|---|
| use MSE instead of dot-product inner loss (ablation) | DeiT-S TTT · ImageNet-1K val | top-1 accuracy: 78.9 -> 79.2 | No clear effect |
| replace inner loss with MAE (L1) (ablation) | DeiT-S TTT · ImageNet-1K val | top-1 accuracy: 78.9 -> 76.5 | Hurt |
| use sequential mini-batch inner updates instead of full batch (ablation) | DeiT-S TTT · ImageNet-1K val | top-1 accuracy: 78.9 -> 78.1 | Hurt |
| increase inner epochs from 1 to 3 (ablation) | DeiT-S TTT · ImageNet-1K val | top-1 accuracy: 78.9 -> 79.2; FPS 1315 -> 787 | Mixed |
| inner learning rate 1.0 versus 0.1 (ablation) | DeiT-S TTT · ImageNet-1K val | top-1 accuracy: 77.5 -> 78.9 | Improved |
| very large inner learning rate (5.0, 10.0) (ablation) | DeiT-S TTT · ImageNet-1K val | top-1 accuracy: 78.9 -> 76.7* | Hurt |
| token-wise dynamic inner learning rate (ablation) | DeiT-S TTT · ImageNet-1K val | top-1 accuracy: 78.9 -> 78.7 | Hurt |
| widen inner MLP hidden ratio from 1 to 4 (ablation) | DeiT-S TTT · ImageNet-1K val | top-1 accuracy: 78.9 -> 79.6 | Improved |
| deepen inner MLP from 2 to 3 layers (ablation) | DeiT-S TTT · ImageNet-1K val | top-1 accuracy: 78.9 -> 77.5 | Hurt |
| remove output layer of inner MLP, SiLU(FC(x)) (ablation) | DeiT-S TTT · ImageNet-1K val | top-1 accuracy: 78.9 -> 79.4 | Improved |
| add residual connection to inner MLP (ablation) | DeiT-S TTT · ImageNet-1K val | top-1 accuracy: 78.9 -> 78.8 | No clear effect |
| use depthwise conv inner model instead of 2-layer MLP (ablation) | DeiT-S TTT · ImageNet-1K val | top-1 accuracy: 78.9 -> 80.1 | Improved |
Closest papers by meaning
- The Illusion of Progress? A Critical Look at Test-Time Adaptation for Vision-Language Models (NeurIPS 2025 · similarity 0.89)
- Test-Time Visual In-Context Tuning (CVPR 2025 · similarity 0.89)
- Diversity Matters: Revisiting Test-Time Compute in Vision-Language Models (ICML 2026 · similarity 0.89)
- AVA-Bench: Atomic Visual Ability Benchmark for Vision Foundation Models (CVPR 2026 · similarity 0.88)
- Learning to See Before Seeing: Demystifying LLM Visual Priors from Language Pre-training (ICLR 2026 · similarity 0.88)
- TTRV: Test-Time Reinforcement Learning for Vision Language Models (CVPR 2026 · similarity 0.88)
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.
