Research map / Vision-language and multimodal / Multimodal LLMs
Q-Bench+: A Benchmark for Multi-Modal Foundation Models on Low-Level Vision From Single Images to Pairs
Zicheng Zhang, Haoning Wu, Erli Zhang, Guangtao Zhai, Weisi Lin
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 |
|---|---|---|---|
| use softmax probability of good vs poor instead of argmax (ablation) | LLaVA-v1 (Vicuna-13B) · KonIQ-10k | SRCC: 0.038 -> 0.462 | Improved |
| use softmax probability of good vs poor instead of argmax (ablation) | mPLUG-Owl (LLaMA-7B) · KonIQ-10k | SRCC: 0.111 -> 0.409 | Improved |
| use softmax probability of good vs poor instead of argmax (ablation) | Shikra (Vicuna-7B) · KonIQ-10k | SRCC: 0.178 -> 0.314 | Improved |
| use softmax probability of good vs poor instead of argmax (ablation) | InstructBLIP (Vicuna-7B) · CGIQA-6K | SRCC: 0.141 -> 0.263 | Improved |
| add prompt ensemble (good+high+fine vs poor+low+bad) (ablation) | Emu2-Chat · IQA average | SRCC: 0.593 -> 0.625 | Improved |
| add prompt ensemble (good+high vs poor+low) (ablation) | Qwen-VL · IQA average | SRCC: 0.475 -> 0.495 | Improved |
| add prompt ensemble (good+high+fine vs poor+low+bad) (ablation) | LLaVA-v1.5 (Vicuna-v1.5-7B) · IQA average | SRCC: 0.424 -> 0.439 | Improved |
| add prompt ensemble (good+fine vs poor+bad) (ablation) | InternLM-XComposer-VL · IQA average | SRCC: 0.541 -> 0.543 | No clear effect |
| fine-tune mPLUG-Owl2 on single-image low-level data (Q-Instruct) | mPLUG-Owl2 · perception-pair test | overall accuracy: 48.94% -> 53.15% | Improved |
| fine-tune mPLUG-Owl2 on single-image low-level data (Q-Instruct) | mPLUG-Owl2 · description-pair | precision score: 0.92 -> 0.79 | Hurt |
| go from single images to image pairs | junior-level human · perception | overall accuracy: 74.31% -> 80.12% | Improved |
| go from single images to image pairs | GPT-4V · perception | overall accuracy: 74.10% -> 78.07% | Improved |
Closest papers by meaning
- Q-Bench: A Benchmark for General-Purpose Foundation Models on Low-level Vision (ICLR 2024 · similarity 0.98)
- Q-Instruct: Improving Low-level Visual Abilities for Multi-modality Foundation Models (CVPR 2024 · similarity 0.94)
- Are We on the Right Way for Evaluating Large Vision-Language Models? (NeurIPS 2024 · similarity 0.92)
- Benchmarking Large Vision-Language Models on Fine-Grained Image Tasks: A Comprehensive Evaluation (ICLR 2026 · similarity 0.92)
- II-Bench: An Image Implication Understanding Benchmark for Multimodal Large Language Models (NeurIPS 2024 · similarity 0.91)
- MLLMs Know Where to Look: Training-free Perception of Small Visual Details with Multimodal LLMs (ICLR 2025 · similarity 0.91)
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.
