Research map / Vision-language and multimodal / Multimodal LLMs
LVLM-EHub: A Comprehensive Evaluation Benchmark for Large Vision-Language Models
Peng Xu, Wenqi Shao, Kaipeng Zhang, Peng Gao, Shuo Liu, Meng Lei, Fanqing Meng, Siyuan Huang, Yu Qiao, Ping Luo
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 multi-turn reasoning evaluation instead of single-turn QA on SNLI-VE | instruction-tuned LVLMs vs BLIP2 · SNLI-VE | accuracy: BLIP2 32.00 vs LLaVA 57.80, InstructBLIP 59.00 | Improved |
| multi-turn reasoning evaluation on VCR | instruction-tuned LVLMs vs BLIP2 · VCR | accuracy: BLIP2 31.60 vs InstructBLIP 54.20 | Improved |
| massive VQA instruction tuning (InstructBLIP) vs BLIP2 | InstructBLIP · visual perception, knowledge, reasoning | average score: 0.858 -> 0.928 perception; 0.927 -> 0.967 knowledge acquisition | Improved |
| add massive in-domain VQA tuning (InstructBLIP) on embodied planning | InstructBLIP · VirtualHome | average score: 0.922 (LA-V2) vs 0.772 (InstructBLIP) | Hurt |
| instruction tuning with moderate instruction data on POPE hallucination | LLaVA, LA-V2, mPLUG-Owl, Otter vs BLIP2 · MSCOCO-Random POPE | accuracy: 82.21 (BLIP2) -> 51.52 (LLaVA), 74.44 (LA-V2) | Hurt |
| change the prompt for image captioning | VPGTrans · NoCaps | CIDEr: 19.66 to 48.13 | Mixed |
| add visual instruction tuning (vs BLIP2) for embodied planning | LLaMA-Adapter-v2, LLaVA · VirtualHome | executability: 2.88 -> 4.08 (LA-V2), 3.82 (LLaVA) | Improved |
| add image-text pair pretraining (204M pairs) to instruction tuning | mPLUG-Owl vs Otter · VirtualHome | spatial relation score: 3.10 -> 3.22 | Improved |
| larger visual encoder and Q-Former trained on image-text pairs | BLIP2, InstructBLIP, VPGTrans · OCR/KIE/captioning | average score: 0.927, 0.967, 0.720 vs LLaVA 0.377 | Improved |
| update LLM with 158K instruction-following data | mPLUG-Owl, LLaVA · ImageNet1K, CIFAR10 | accuracy: 26.68 (mPLUG-Owl ImageNet1K), 67.96 (LLaVA CIFAR10) | Improved |
Closest papers by meaning
- Are We on the Right Way for Evaluating Large Vision-Language Models? (NeurIPS 2024 · similarity 0.93)
- Dynamic Multimodal Evaluation with Flexible Complexity by Vision-Language Bootstrapping (ICLR 2025 · similarity 0.93)
- Benchmarking Large Vision-Language Models on Fine-Grained Image Tasks: A Comprehensive Evaluation (ICLR 2026 · similarity 0.91)
- MMT-Bench: A Comprehensive Multimodal Benchmark for Evaluating Large Vision-Language Models Towards Multitask AGI (ICML 2024 · similarity 0.91)
- MMIE: Massive Multimodal Interleaved Comprehension Benchmark for Large Vision-Language Models (ICLR 2025 · similarity 0.91)
- MMDU: A Multi-Turn Multi-Image Dialog Understanding Benchmark and Instruction-Tuning Dataset for LVLMs (NeurIPS 2024 · 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.
