Research map / Vision-language and multimodal / CLIP and zero-shot
Going Deep: Deep Visual Prompting with LoTeP
Zijie Zhao, Yanru Wu, Yuji Wang, Haohua Wang, Enming Zhang, Wai Kin Chan, Yang Li
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 |
|---|---|---|---|
| model deep prompts as low-rank tensors with layerwise rank decay instead of input-space low-rank VP | LoR-VP · 10 datasets average (ViT-B/32) | accuracy: 88.94 -> 91.44 | Improved |
| same change on ResNet-18 | LoR-VP · 10 datasets average (ResNet-18) | accuracy: 78.15 -> 82.19 | Improved |
| use shared prompts across grouped channels in deep VP (intermediate group size) (ablation) | deep LoR-VP · CIFAR-100 / Tiny-ImageNet | accuracy | Improved |
| apply LoTeP instead of LoR-VP for out-of-distribution transfer | LoR-VP · ImageNet-A (ConvNeXt-B) | accuracy: 39.36 -> 44.41 | Improved |
| apply LoTeP on small-capacity ResNet-18 versus full fine-tuning | LoTeP-TK · 10 datasets average (ResNet-18) | accuracy: 84.62 -> 82.19 | Hurt |
| use cosine rank decay instead of constant rank (ablation) | LoTeP · CIFAR-100 (CoAtNet-2) | accuracy | Improved |
| use cosine rank decay instead of linear decay (ablation) | LoTeP · CIFAR-100 (CoAtNet-2) | accuracy | Improved |
| increase prompting depth ratio rho (ablation) | LoTeP-TK · CIFAR-100 (ResNet-50 vs Swin-B) | accuracy | Mixed |
| increase input rank r0 (ablation) | LoTeP-CP · CIFAR-100 (ResNet-18, CoAtNet-2) | accuracy | Mixed |
| use LoTeP rather than LoR-VP for efficiency | LoR-VP · Tiny-ImageNet (ViT-B/16) | training time: 1.65h -> 1.57h | Improved |
| use TK instead of CP variant (ablation) | LoTeP · CIFAR-100 across nine backbones | accuracy | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- LoR-VP — LoR-VP: Low-Rank Visual Prompting for Efficient Vision Model Adaptation (ICLR 2025) · better on 10 of 11 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- VIPAMIN: Visual Prompt Initialization via Embedding Selection and Subspace Expansion (NeurIPS 2025 · similarity 0.90)
- LoR-VP: Low-Rank Visual Prompting for Efficient Vision Model Adaptation (ICLR 2025 · similarity 0.89)
- 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)
- Revisiting the Power of Prompt for Visual Tuning (ICML 2024 · similarity 0.88)
- SA²VP: Spatially Aligned-and-Adapted Visual Prompt (AAAI 2024 · 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.
