Research map / Vision-language and multimodal / CLIP and zero-shot
Interpretable Zero-Shot Learning with Locally-Aligned Vision-Language Model
Shiming Chen, Bowen Duan, Salman Khan, Fahad Shahbaz Khan
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 OT local visual-semantic alignment to DCLIP baseline (ablation) | DCLIP (CLIP ViT-B/16) · ImageNet | accuracy: 67.9 -> 68.5 | Improved |
| add OT local visual-semantic alignment to DCLIP baseline (ablation) | DCLIP (CLIP ViT-B/16) · CUB | accuracy: 57.8 -> 59.0 | Improved |
| add OT local visual-semantic alignment to DCLIP baseline (ablation) | DCLIP (CLIP ViT-B/16) · Place365 | accuracy: 40.3 -> 41.6 | Improved |
| add vision selection on top of OT (ablation) | Baseline+OT (CLIP ViT-B/16) · CUB | accuracy: 59.0 -> 60.0 | Improved |
| add vision selection on top of OT (ablation) | Baseline+OT (CLIP ViT-B/16) · ImageNet | accuracy: 68.5 -> 69.0 | Improved |
| add region-global hybrid cost on top of OT (ablation) | Baseline+OT (CLIP ViT-B/16) · CUB | accuracy: 59.0 -> 59.3 | Improved |
| add region-global hybrid cost on top of OT (ablation) | Baseline+OT (CLIP ViT-B/16) · Place365 | accuracy: 41.6 -> 41.9 | Improved |
| add hybrid cost to OT+VS (full LaZSL) (ablation) | Baseline+OT+VS (CLIP ViT-B/16) · CUB | accuracy: 60.0 -> 60.3 | Improved |
| LaZSL versus DCLIP across backbones | DCLIP · five-dataset average | accuracy: +2.0 (B/32), +1.6 (B/16), +1.2 (L/14) | Improved |
| LaZSL versus CuPL on ImageNet and Oxford Pets with ViT-B/16 | CuPL · ImageNet | accuracy: 69.6 -> 69.2 | Hurt |
| set hybrid coefficient theta too high (>=0.9) (ablation) | LaZSL · CUB and ImageNet | accuracy | Hurt |
| set cropping scale alpha too large (>=0.6) (ablation) | LaZSL · CUB and ImageNet | accuracy | Hurt |
Closest papers by meaning
- Improved Zero-Shot Classification by Adapting VLMs with Text Descriptions (CVPR 2024 · similarity 0.90)
- Visual-Text Cross Alignment: Refining the Similarity Score in Vision-Language Models (ICML 2024 · similarity 0.89)
- ZeroDiff: Solidified Visual-semantic Correlation in Zero-Shot Learning (ICLR 2025 · similarity 0.89)
- From Local Details to Global Context: Advancing Vision-Language Models with Attention-Based Selection (ICML 2025 · similarity 0.89)
- Locality Alignment Improves Vision-Language Models (ICLR 2025 · similarity 0.89)
- Progressive Semantic-Guided Vision Transformer for Zero-Shot Learning (CVPR 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.
