Research map / Vision-language and multimodal / CLIP and zero-shot
VITAL: Vision-Encoder-centered Pre-training for LMMs in Visual Quality Assessment
Ziheng Jia, Linhan Cao, Jinliang Han, Zicheng Zhang, Jiaying Qian, Jiarui Wang, Zijian Chen, Guangtao Zhai, Xiongkuo Min
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 PMOD (machine opinion distribution) prediction instead of CE on mean level (ablation) | VITAL-Base-8B · KADID | SRCC: 0.602 -> 0.759 | Improved |
| add PMOD prediction (ablation) | VITAL-Base-8B · LIVE-VQC | SRCC: 0.725 -> 0.800 | Improved |
| add pairwise preference training (ablation) | VITAL-Base-8B · KADID | SRCC: 0.725 -> 0.759 | Improved |
| add text generation tasks to pre-training (ablation) | VITAL-Base-8B · YT-Gaming | SRCC: 0.712 -> 0.710 | No clear effect |
| add text generation tasks to pre-training (ablation) | VITAL-Base-8B · KADID | SRCC: 0.743 -> 0.759 | Improved |
| vision-encoder-centered pre-training on machine-annotated data vs SFT on human labels | InternVL3-8B · video scoring avg (8 datasets) | AVG SRCC/PLCC: 0.758 -> 0.820 | Improved |
| pair VITAL encoder with new decoder without any training (Zero) vs original instruct model | InternVL3-1B · video scoring avg (8 datasets) | AVG SRCC/PLCC: 0.405 -> 0.611 | Improved |
| warm-up decoder with 4000 samples after transfer | VITAL-Zero-1B · video scoring avg (8 datasets) | AVG SRCC/PLCC: 0.611 -> 0.808 | Improved |
| transfer decoder to 14B with warm-up | VITAL-Base-8B · video scoring avg (8 datasets) | AVG SRCC/PLCC: 0.820 -> 0.795 | Mixed |
| post-train with VITAL pre-training before SFT | VITAL-Assistant-8B · QBench-video-test-single | overall accuracy (%): 42.67% -> 62.94% | Improved |
| use focal loss instead of CE in post-training (ablation) | VITAL-Assistant-8B · QBench-video open-ended questions (316) | output length and accuracy | Improved |
| linear probe on VITAL encoder vs Simple-VQA | VITAL Vision Encoder · LIVE-VQC | SRCC: 0.749 -> 0.760 | Improved |
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.
- Q-Align — Q-Align: Teaching LMMs for Visual Scoring via Discrete Text-Defined Levels (ICML 2024) · better on 1 of 1 shared result
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
- Adaptive Image Quality Assessment via Teaching Large Multimodal Model to Compare (NeurIPS 2024 · similarity 0.89)
- Are We on the Right Way for Evaluating Large Vision-Language Models? (NeurIPS 2024 · similarity 0.89)
- VLM-RobustBench: A Comprehensive Benchmark for Robustness of Vision-Language Models (ICML 2026 · similarity 0.88)
- Beyond Encoder Accumulation: Measuring Encoder Roles in Multi-Encoder VLMs (NeurIPS 2026 · similarity 0.88)
- Self-Evolving Vision-Language Models for Image Quality Assessment via Voting and Ranking (ICLR 2026 · similarity 0.88)
- LL-Bench: Rethinking Low-Level Vision Evaluation in the Era of Large-Scale Generative Models (NeurIPS 2026 · similarity 0.88)
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.
