AI Paper Atlas

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

Toward a Holistic Evaluation of Robustness in CLIP Models

Weijie Tu, Weijian Deng, Tom Gedeon

TPAMI 2025 · Vision-language and multimodal › CLIP and zero-shot · approach: Fine-grained improvements to CLIP · 20 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
fine-tune on ImageNet-1K with cross-entropyzero-shot CLIP ViT-B/32 · cue-conflictshape bias: 0.575 -> 0.401Hurt
fine-tune on ImageNet-1K with cross-entropyzero-shot CLIP ViT-B/16 · cue-conflictshape bias: 0.473 -> 0.345Hurt
fine-tune with contrastive loss instead of cross-entropy (ablation)CLIP ViT-B/32 · cue-conflictshape bias: 0.401 -> 0.561Improved
fine-tune with contrastive loss instead of cross-entropy (ablation)CLIP ViT-B/16 · cue-conflictshape bias: 0.345 -> 0.448Improved
increase input resolution during fine-tuning (ablation)CLIP · ImageNet val vs Stylized-ImageNetaccuracyMixed
additional fine-tuning on ImageNet-12K before ImageNet-1K (ablation)CLIP-FT · OOD detectionAUROCImproved
fine-tune CLIP on ImageNetCLIP · calibrationECEHurt
apply temperature scalingzero-shot CLIP · ImageNet and shiftsNLL/ECEImproved
use a single prompt instead of a large prompt set (ablation)CLIP WIT models · classification, OOD, calibrationaccuracyMixed
add LLM-generated prompt sets (VDT, MPVR, REAL) (ablation)CLIP WIT models · classificationaccuracyImproved
swap CLIP for LLaVA with same visual encoder when category list is hard for CLIPCLIP ViT-L/14-336 · ImageNet-D Backgroundaccuracy: 0.14 -> 0.57Improved
swap CLIP for LLaVA with same visual encoder when category list is easy for CLIPCLIP ViT-L/14-336 · ImageNet-D Backgroundaccuracy: 0.90 -> 0.92No clear effect

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