Research map / Label-efficient and robust learning / Test-time adaptation
Convolutional Visual Prompt for Robust Visual Perception
Yun-Yun Tsai, Chengzhi Mao, Junfeng Yang
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 unstructured visual prompting with a random-initialized 3×3 convolutional prompt | WideResNet18 · CIFAR-10-C | average error (%): 58.24 -> 52.37; -5.87 | Improved |
| Replace unstructured visual prompting with a random-initialized 3×3 convolutional prompt | ResNet26 · CIFAR-10-C | average error (%): 62.12 -> 59.32; -2.80 | Improved |
| Apply CVP with a random-initialized 3×3 kernel | ResNet50 · ImageNet-C | mCE: 76.87 -> 75.34; -1.49 | Improved |
| Apply CVP with a random-initialized 5×5 kernel | ResNet50 · ImageNet-R | error (%): 63.83 -> 63.06; -0.77 | Improved |
| Apply CVP with a random-initialized 3×3 kernel | ResNet50 · ImageNet-Sketch | error (%): 75.90 -> 75.30; -0.60 | Improved |
| Apply CVP with a random-initialized 5×5 kernel | ResNet50 · ImageNet-A | error (%): 100.0 -> 98.4; -1.6 | Improved |
| Use a convolutional prompt instead of a rank-3 low-rank visual prompt (ablation) | Visual prompting · CIFAR-10-C | average error (%): 54.86 -> 52.37; -3.38 vs. standard for LVP, -5.87 for CVP | Improved |
| Use unstructured rather than convolutional prompting with supervised labels (ablation) | Visual prompting · CIFAR-10-C | average error (%): 49.59 -> 42.02 | Improved |
| Add CVP input adaptation on top of TENT weight adaptation (ablation) | TENT · CIFAR-10-C | average error (%): 38.52 -> 36.69; -1.83 | Improved |
| Add CVP input adaptation on top of TTT (ablation) | TTT · CIFAR-10-C | average error (%): 52.92 -> 53.07 | Hurt |
| Use MAE rather than contrastive learning as the self-supervised objective (ablation) | CVP · ImageNet-C, severity 1 | robust accuracy: 58.97 -> 60.94 | Improved |
| Disable test-time kernel updates (ablation) | CVP with random 3×3 kernel · CIFAR-10-C | average error (%): 53.77 -> 52.37 | Hurt |
Closest papers by meaning
- AutoVP: An Automated Visual Prompting Framework and Benchmark (ICLR 2024 · similarity 0.90)
- Understanding Model Reprogramming for CLIP via Decoupling Visual Prompts (ICML 2025 · similarity 0.90)
- Enhancing Visual Prompting through Expanded Transformation Space and Overfitting Mitigation (NeurIPS 2025 · similarity 0.89)
- LaViP: Language-Grounded Visual Prompting (AAAI 2024 · similarity 0.89)
- LoR-VP: Low-Rank Visual Prompting for Efficient Vision Model Adaptation (ICLR 2025 · similarity 0.89)
- SA²VP: Spatially Aligned-and-Adapted Visual Prompt (AAAI 2024 · similarity 0.89)
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.
