AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

A Visual Leap in CLIP Compositionality Reasoning through Generation of Counterfactual Sets

Zexi Jia, Chuanwei Huang, Hongyan Fei, Yeshuang Zhu, Zhiqiang Yuan, Ying Deng, Jiapei Zhang, Jinchao Zhang, Jie Zhou

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Chain-of-thought multimodal reasoning · 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
replace contrastive loss with set-based paired loss L_sets (ablation)CLIP ViT-B/32 · ARO VG-Relationaccuracy: 62.8 -> 82.3Improved
add stitched real-data images to training (ablation)CLIP ViT-B/32 · ARO Order (COCO)accuracy: 76.1 -> 90.8Improved
add block-diffusion generated counterfactual images (ablation)CLIP ViT-B/32 · ARO Order (Flickr)accuracy: 87.8 -> 93.0Improved
add negative-text loss L_neg (ablation)CLIP ViT-B/32 · WinogroundText score: 21.5 -> 32.6Improved
increase counterfactual set size from 5 to 20 (ablation)CLIP ViT-B/32 · ARO VG-Relationaccuracy: 84.9 -> 85.7Improved
combine local image and local text guidance in block-based diffusion (ablation)counterfactual generation · counterfactual data qualityrelation accuracy (%): 76.9 -> 95.2Improved
scale generated data from 10k to 300k (ablation)GCS fine-tuned CLIP · VL-Checklist Relationaccuracy: 74.5 -> 85.6Improved
fine-tune with GCS instead of text-perturbation methodsCLIP · WinogroundGroup score: 8.0 (NegCLIP) -> 10.0Improved
fine-tune with GCSCLIP · Elevater all-shotaccuracy: 78.96 -> 78.50No clear effect
use proposed loss instead of sigmoid loss (ablation)CLIP fine-tuning · training time on 4 V100training time reduction: 16.2% less than sigmoid loss, 13.6% less than contrastiveImproved
fine-tune BLIP-2 with GCS-300kBLIP-2 · ARO-Raccuracy: 41.2 -> 74.9Improved
fine-tune MiniGPT-4 with GCS-300kMiniGPT-4 · Winogroundaccuracy: 8.3 -> 15.3Improved

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