Research map / Vision-language and multimodal / Multimodal LLMs
Q-REAL: Towards Naturalness and Distortion Evaluation for AI-Generated Content
Shushi Wang, Zicheng Zhang, Chunyi Li, Wei Wang, Liya Ma, Xiaoyu Li, Fengjiao Chen, Xuezhi Cao, Guangtao Zhai, Xiaohong Liu
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 |
|---|---|---|---|
| fine-tune on Q-Real (Stages 1-2) instead of zero-shot | Qwen2.5-VL-7B-Instruct · ImageQA Distortion | detection rate: 0.059 -> 0.548 | Improved |
| fine-tune on Q-Real instead of zero-shot | Qwen3-VL-8B-Instruct · ImageQA Naturalness | detection rate: 0.313 -> 0.548 | Improved |
| fine-tune on Q-Real instead of zero-shot | Qwen2.5-VL-7B-Instruct · ObjectQA Naturalness Q1 | accuracy: 0.605 -> 0.751 | Improved |
| fine-tune on Q-Real instead of zero-shot | InternVL3-8B · ImageQA Naturalness | LLM score: 0.230 -> 0.737 | Improved |
| fine-tune on Q-Real instead of zero-shot | Qwen3-VL-8B-Instruct · ImageQA Naturalness | IoU: 0.848 -> 0.834 | No clear effect |
| add Stage 1 object-level judgment training before Stage 2 (ablation) | Qwen2.5-VL-7B-Instruct · ImageQA Distortion | LLM score: 0.623 -> 0.633 | Improved |
| add Stage 1 object-level judgment training before Stage 2 (ablation) | InternVL2.5-8B · ImageQA Distortion | detection rate: 0.418 -> 0.448 | Improved |
| add Stage 1 object-level judgment training before Stage 2 (ablation) | Qwen3-VL-8B-Instruct · ImageQA Naturalness | LLM score: 0.721 -> 0.722 | No clear effect |
| add training-inference consistency to Stage 3 prompt (ablation) | Q-Real-Score · Q-Real Bench quality scoring | SRCC: 0.692 -> 0.710 | Improved |
| add naturalness CoT issues to Stage 3 prompt (ablation) | Q-Real-Score · Q-Real Bench quality scoring | SRCC: 0.700 -> 0.710 | Improved |
| add distortion CoT issues to Stage 3 prompt (ablation) | Q-Real-Score · Q-Real Bench quality scoring | SRCC: 0.696 -> 0.710 | Improved |
| use detailed issue descriptions in prompt (Q-Real-Score) instead of Q-Eval-Score general CoT | Qwen3-VL-8B · AGIQA-3K | SRCC: 0.587 -> 0.608 | 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
- A-Bench: Are LMMs Masters at Evaluating AI-generated Images? (ICLR 2025 · similarity 0.88)
- Revisiting MLLM Based Image Quality Assessment: Errors and Remedy (AAAI 2026 · similarity 0.88)
- Teaching Large Language Models to Regress Accurate Image Quality Scores Using Score Distribution (CVPR 2025 · similarity 0.87)
- Grounding-IQA: Grounding Multimodal Language Model for Image Quality Assessment (ICLR 2026 · similarity 0.87)
- AI-generated Image Quality Assessment in Visual Communication (AAAI 2025 · similarity 0.85)
- AIGV-Assessor: Benchmarking and Evaluating the Perceptual Quality of Text-to-Video Generation with LMM (CVPR 2025 · similarity 0.85)
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.
