Research map / Vision-language and multimodal / CLIP and zero-shot
Is CLIP ideal? No. Can we fix it? Yes!
Raphi Kang, Yue Song, Georgia Gkioxari, Pietro Perona
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 functional rows (constant vectors for spatial words) (ablation) | DCSM (synthetic-trained) · WhatsUp | accuracy: 48.6 -> 62.6 | Improved |
| add functional rows (constant vectors for spatial words) (ablation) | DCSM (synthetic-trained) · COCO 2-obj | accuracy: 55.5 -> 70.9 | Improved |
| use a 2-layer CNN scorer instead of a small ViT (ablation) | DCSM (synthetic-trained) · WhatsUp | accuracy: 29.4 -> 62.6 | Improved |
| use a 2-layer CNN scorer instead of a small ViT (ablation) | DCSM (synthetic-trained) · COCO 2-obj | accuracy: 47.0 -> 70.9 | Improved |
| finetune CLIP on the synthetic data (instead of DCSM) (ablation) | CLIP ViT-B/16 · WhatsUp | accuracy: 30.5 -> 25.5 | Hurt |
| train an MLP scorer on CLIP embeddings (ablation) | CLIP ViT-B/16 · WhatsUp | accuracy: 30.5 -> 25.4 | Hurt |
| train an MLP scorer on CLIP embeddings (ablation) | CLIP ViT-B/16 · COCO 2-obj | accuracy: 45.9 -> 53.6 | Mixed |
| replace cosine similarity with DCSM scoring (COCO-trained) | CLIP ViT-B/16 · CLEVR-bind | accuracy: 20.2 -> 39.9 | Improved |
| replace cosine similarity with DCSM scoring (COCO-trained) | CLIP ViT-B/16 · WhatsUp | accuracy: 30.5 -> 63.7 | Improved |
| replace cosine similarity with DCSM scoring (COCO-trained) | CLIP ViT-B/16 · VG_attr | accuracy: 61.8 -> 68.1 (NegCLIP 72.4) | Mixed |
| train DCSM on synthetic instead of COCO data | DCSM · NegBench COCO | accuracy: 48.6 -> 38.8 | Hurt |
| add LLM-updated functional-word lookup (open vocab) | DCSM · VL-Checklist VG spatial | accuracy: 50.6 -> 63.5 | Improved |
Closest papers by meaning
- Diffusion Feedback Helps CLIP See Better (ICLR 2025 · similarity 0.92)
- Demystifying CLIP Data (ICLR 2024 · similarity 0.91)
- un$^2$CLIP: Improving CLIP's Visual Detail Capturing Ability via Inverting unCLIP (NeurIPS 2025 · similarity 0.91)
- Meta CLIP 2: A Worldwide Scaling Recipe (NeurIPS 2025 · similarity 0.90)
- CLIPLoss and Norm-Based Data Selection Methods for Multimodal Contrastive Learning (NeurIPS 2024 · similarity 0.90)
- Unlocking Fine-Grained Perception in CLIP via Structurally-Aware Latent Masked Modeling (NeurIPS 2026 · 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.
