Research map / Vision-language and multimodal / CLIP and zero-shot
A Closer Look at the Robustness of Contrastive Language-Image Pre-Training (CLIP)
Weijie Tu, Weijian Deng, Tom Gedeon
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 |
|---|---|---|---|
| fine-tune CLIP with larger input resolution (ablation) | CLIP-FT (LAION, pair 1) · cue-conflict | shape bias: 0.51 -> 0.42 | Hurt |
| fine-tune CLIP with larger input resolution (ablation) | CLIP-FT (LAION, pair 1) · Stylized-ImageNet | SIN accuracy: 0.32 -> 0.28 | Hurt |
| fine-tune CLIP with larger input resolution (ablation) | CLIP-FT (LAION, pair 1) · ImageNet-Val | accuracy: 0.88 -> 0.89 | Improved |
| fine-tune CLIP with larger input resolution (ablation) | CLIP-FT (LAION, pair 4) · ImageNet-Val | accuracy: 0.83 -> 0.85 | Improved |
| fine-tune CLIP with larger input resolution (ablation) | CLIP-FT (WIT, pair 1) · ImageNet-Val | accuracy: 0.88 -> 0.88 | No clear effect |
| fine-tune CLIP on ImageNet-1K | zero-shot CLIP · cue-conflict | shape bias | Hurt |
| scale up CLIP model (ViT-L/14 vs ViT-B/32, LAION-80M) | CLIP · cue-conflict | shape bias: 0.09 lower | Hurt |
| additionally fine-tune CLIP on ImageNet-12K before ImageNet-1K | CLIP-FT · OOD detection (5 scenarios) | AUROC/FPR@95 | Improved |
| fine-tune CLIP on ImageNet-1K | zero-shot CLIP · OOD detection | AUROC/FPR@95 | Mixed |
| train CLIP on WIT instead of LAION | CLIP · OOD detection (three scenarios) | AUROC | Improved |
| fine-tune CLIP on ImageNet-1K | zero-shot CLIP · calibration | ECE | Improved |
| apply temperature scaling on ID calibration set | CLIP · ImageNet-A | calibration error | Improved |
Closest papers by meaning
- Post-pre-training for Modality Alignment in Vision-Language Foundation Models (CVPR 2025 · similarity 0.91)
- A Hard-to-Beat Baseline for Training-free CLIP-based Adaptation (ICLR 2024 · similarity 0.91)
- Demystifying CLIP Data (ICLR 2024 · similarity 0.90)
- CLIPTTA: Robust Contrastive Vision-Language Test-Time Adaptation (NeurIPS 2025 · similarity 0.90)
- Calibrating Multi-modal Representations: A Pursuit of Group Robustness without Annotations (CVPR 2024 · similarity 0.90)
- Diffusion Feedback Helps CLIP See Better (ICLR 2025 · similarity 0.90)
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.
