AI Paper Atlas

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

CLIP Behaves like a Bag-of-Words Model Cross-modally but not Uni-modally

Darina Koishigarina, Arnas Uselis, Seong Joon Oh

ICLR 2026 · Vision-language and multimodal › CLIP and zero-shot · approach: CLIP training recipes and data curation · 23 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 hard negative captions to the training batch (ablation)LABCLIP · PUG:SPARE (ViT-L/14)test accuracy: 0.90 -> 0.94Improved
add hard negative captions to the training batch (ablation)LABCLIP · PUG:SPARE (ViT-B/32)test accuracy: 0.89 -> 0.93Improved
add hard negative captions to the training batch (ablation)LABCLIP · AROaccuracyImproved
add a learned linear map on text embeddingsCLIP · CLEVRcross-modal test accuracy: 0.58 -> 0.95Improved
add a learned linear map on text embeddingsCLIP · PUG:SPARcross-modal test accuracy: 0.53 -> 0.97Improved
add a learned linear map on text embeddingsCLIP · PUG:SPAREcross-modal test accuracy: 0.50 -> 0.94Improved
add a learned linear map on text embeddingsCLIP ViT-B/32 · COCO retrievalRecall@1: 0.30 -> 0.41Improved
add a learned linear map on text embeddingsCLIP ViT-B/32 · ARO COCO PRCaccuracy: 0.48 -> 0.81Improved
align the text space to the image space (cosine similarity of probe weights)CLIP · PUG:SPAREprobe weight cosine similarity: 0.09 -> 0.65Improved
train the linear map on CC3M instead of COCO (ablation)LABCLIP · ARO COCO PRCaccuracy: 0.81 -> 0.73Hurt
apply the linear map to image embeddings instead of text embeddings (ablation)LABCLIP · CLEVRcross-modal test accuracy: 0.95 -> 0.98No clear effect
use BoW-trained encoders instead of pretrained CLIP embeddings (ablation)linear probe · CLEVR image embeddingsprobe test accuracy: 0.96 -> 0.66Hurt

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