AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Unsupervised Visual Chain-of-Thought Reasoning via Preference Optimization

Kesen Zhao, Beier Zhu, Qianru Sun, Hanwang Zhang

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Chain-of-thought multimodal reasoning · 19 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 image-level CoT inference with generated bounding boxes (ablation)UV-CoT · Six VQA benchmarks averageGPT-4o score (0-1): 0.417 -> 0.494Improved
use ground-truth bounding boxes instead of predicted ones (ablation)UV-CoT · Six VQA benchmarks averageGPT-4o score (0-1): 0.494 -> 0.618Improved
replace naive DPO with Score-DPO (sDPO) (ablation)UV-CoT · Six VQA benchmarks averageGPT-4o score (0-1): 0.475 -> 0.494Improved
add iterative learning (ablation)UV-CoT · Six VQA benchmarks averageGPT-4o score (0-1): 0.459 -> 0.494Improved
add next-response term gamma in evaluation (ablation)UV-CoT · Six VQA benchmarks averageGPT-4o score (0-1): 0.406 -> 0.494Improved
use OmniLMM-12B evaluator instead of self-evaluation (ablation)UV-CoT · Six VQA benchmarks averageGPT-4o score (0-1): 0.441 -> 0.473Improved
add UV-CoT bounding-box CoT at inferenceOmniLMM-12B · Six VQA benchmarks averageGPT-4o score (0-1): 0.443 -> 0.516Improved
add UV-CoT bounding-box CoT at inferenceLLaVA-1.5-7B · Six VQA benchmarks averageGPT-4o score (0-1): 0.393 -> 0.440Improved
add 10% labeled SFT before preference optimizationUV-CoT · Six VQA benchmarks averageGPT-4o score (0-1): 0.473 -> 0.494Improved
use unsupervised UV-CoT instead of supervised Visual-CoTLLaVA-1.5-7B · Zero-shot DUDE/SROIE/Visual7w averageGPT-4o score (0-1): 0.261 -> 0.286Improved
train on unlabeled preference data from zero-shot datasets (UV-CoT*)UV-CoT · Zero-shot DUDE/SROIE/Visual7w averageGPT-4o score (0-1): 0.286 -> 0.312Improved
use UV-CoT instead of Visual-CoT-7BLLaVA-1.5-7B · V* Bench zero-shot averageGPT-4o score (0-1): 0.347 -> 0.402Improved

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