AI Paper Atlas

Research map / Vision-language and multimodal / CLIP and zero-shot

Decoupling Template Bias in CLIP: Harnessing Empty Prompts for Enhanced Few-Shot Learning

Zhenyu Zhang, Guangyao Chen, Yixiong Zou, Zhimeng Huang, Yuhua Li

AAAI 2026 · Vision-language and multimodal › CLIP and zero-shot · approach: Fine-grained improvements to CLIP · 5 citations (OpenAlex)

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 changedWhereMeasuredEffect
add template bias correction with empty promptsLoRA-CLIP · 11-dataset average, 1-shottop-1 accuracy: 72.5 -> 73.6Improved
add template bias correction with empty promptsCLIP-LoRA · 11-dataset average, 8-shottop-1 accuracy: 80.3 -> 81.8Improved
add template bias correction with empty promptsLoRA-CLIP · EuroSAT, 1-shottop-1 accuracy: 72.3 -> 78.4Improved
add template bias correction with empty promptsLoRA-CLIP · SUN, 1-shottop-1 accuracy: 70.4 -> 70.5No clear effect
add template bias correction with empty promptsLoRA-CLIP · UCF101, 1-shottop-1 accuracy: 76.3 -> 76.0Hurt
add template bias correction with empty promptsCLIP-LoRA · Food101, 8-shottop-1 accuracy: 83.1 -> 87.3Improved
add template bias correction with empty promptsLoRA-CLIP · Food101, 4-shottop-1 accuracy: 82.7 -> 87.1Improved
add template bias correction with empty promptsLoRA-CLIP · Food101, 1-shottop-1 accuracy: 84.3 -> 85.7Mixed
add template to class name prompt (ablation)CLIP zero-shot · 11-dataset averagetop-1 accuracy: 62.12 -> 64.00Improved
add template to class name prompt (ablation)CLIP zero-shot · EuroSATtop-1 accuracy: 46.38 -> 42.38Hurt
add Ltb loss during fine-tuning (ablation)LoRA CLIP · EuroSAT, 1-shottop-1 accuracy: 72.3 -> 73.9Improved
add pretraining initialization with empty prompts (ablation)LoRA CLIP · EuroSAT, 1-shottop-1 accuracy: 72.3 -> 73.3Improved

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.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy