Research map / Vision-language and multimodal / Multimodal LLMs
BenchLMM: Benchmarking Cross-style Visual Capability of Large Multimodal Models
Rizhao Cai, Zirui Song, Dayan Guan, Zhenhao Chen, Yaohang Li, Xing Luo, Chenyu Yi, Alex Kot
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 Style Prompt Enhancement (predict style before answering) | LLaVA-1.5-13B · BenchLMM artistic styles | average accuracy: 59.3 -> 65.1 | Improved |
| add Style Prompt Enhancement | InstructBLIP-13B · BenchLMM artistic styles | average accuracy: 58.9 -> 62.5 | Improved |
| add Style Prompt Enhancement | LLaVA-1.5-13B · BenchLMM sensor styles | average accuracy: 47.3 -> 51.8 | Improved |
| add Style Prompt Enhancement | InstructBLIP-13B · BenchLMM sensor styles | average accuracy: 37.3 -> 38.7 | Improved |
| add Style Prompt Enhancement | LLaVA-1.5-13B · BenchLMM application styles | average accuracy: 45.2 -> 46.6 | Improved |
| add Style Prompt Enhancement | InstructBLIP-13B · BenchLMM application styles | average accuracy: 38.9 -> 40.3 | Improved |
| add Style Prompt Enhancement | LLaVA-1.5-13B · BenchLMM style-transfer images (DALL-E3) | average accuracy: 57.1 -> 60.8 | Improved |
| add Style Prompt Enhancement | InstructBLIP-13B · BenchLMM style-transfer images (DALL-E3) | average accuracy: 64.0 -> 67.2 | Improved |
| evaluate on non-common styles instead of common photo style | GPT-4V · BenchLMM | accuracy: 81.5 -> 55.0 (Infrared) | Hurt |
| use 13B instead of 7B language model | InstructBLIP · BenchLMM artistic styles | accuracy: Common 73.9 -> 72.7; artistic avg 55.3 -> 58.9 | Mixed |
| SPE applied to application-style data vs artistic/sensor data | SPE · BenchLMM application styles | average accuracy gain: +1.4 vs +5.8 (artistic) | Mixed |
Closest papers by meaning
- ZeroBench: An Impossible Visual Benchmark for Contemporary Large Multimodal Models (ICML 2026 · similarity 0.90)
- LLaVA-Interleave: Tackling Multi-image, Video, and 3D in Large Multimodal Models (ICLR 2025 · similarity 0.89)
- Are We on the Right Way for Evaluating Large Vision-Language Models? (NeurIPS 2024 · similarity 0.88)
- LLaVA-Mini: Efficient Image and Video Large Multimodal Models with One Vision Token (ICLR 2025 · similarity 0.88)
- MMIU: Multimodal Multi-image Understanding for Evaluating Large Vision-Language Models (ICLR 2025 · similarity 0.88)
- Improved Baselines with Visual Instruction Tuning (CVPR 2024 · similarity 0.87)
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.
