AI Paper Atlas

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

HDR-VLM: HDR-Domain Adaptation of VLMs and Preference-Aligned Quality Assessment for HDR Video Color Grading

Hao Yuan, Jiabin Zhang, Yajing Wu, Ruixuan Pang, Jing Li

CVPR 2026 · Vision-language and multimodal › CLIP and zero-shot · approach: Test-time adaptation of VLMs

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 Stage 2 GRPO preference alignment after Stage 1 domain adaptation (ablation)HDR-VLM · HDR color grading datasetPLCC: 0.8024 -> 0.9033Improved
add Stage 1 HDR domain adaptation before RL (ablation)HDR-VLM · HDR color grading datasetPLCC: 0.8140 -> 0.9033Improved
add Stage 1 SFT domain adaptation to a pretrained VLMQwen2.5-VL 7B · HDR color grading datasetPLCC: 0.7828 -> 0.8024Improved
use progressive unfreezing instead of one-shot full-model fine-tuning in Stage 1 (ablation)HDR domain adaptation · HDR domain-adaptation test setaccuracy: 0.85 -> 0.9Improved
unfreeze projection MLP jointly with ViT (ablation)HDR domain adaptation · HDR domain-adaptation test setaccuracy: 0.79 -> 0.82Improved
tune the ViT only on HDR data (ablation)pre-trained VLM · HDR domain-adaptation test setaccuracy: 0.54 -> 0.79Improved
add Step 2 exogenous rewards (RL2 + Rdist) after Step 1 ranking rewards (ablation)HDR-VLM GRPO · HDR color grading datasetPLCC: 0.8860 -> 0.9033Improved
add Step 2 exogenous rewards after Step 1 (ablation)HDR-VLM GRPO · HDR color grading datasetSROCC: 0.8659 -> 0.8667No clear effect
add Step 1 endogenous ranking rewards before Step 2 rewards (ablation)HDR-VLM GRPO · HDR color grading datasetSROCC: 0.7703 -> 0.8667Improved
use HDR-VLM progressive reward instead of VQAThinker rewardVQAThinker · HDR color grading datasetPLCC: 0.8745 -> 0.9033Improved

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