UniPercept is a benchmark dataset for unified perceptual-level image understanding across aesthetics, quality, structure, and texture. Developed by thunderbolt215 and updated in February 2026, it provides a multi-dimensional evaluation framework for multimodal large language models (MLLMs).
Use Cases
- Benchmarking MLLM performance on Image Quality Assessment (IQA) metrics
- Evaluating model sensitivity to Image Aesthetics Assessment (IAA) features
- Testing image understanding across structure and texture (ISTA) variations
Strengths
- Covers four distinct perceptual dimensions: aesthetics, quality, structure, and texture
- Specifically formatted for Multimodal Large Language Model (MLLM) benchmarking
Limitations
- Unknown sample size and record count
- Missing documentation on annotation methodology and source image diversity
Provenance
- Source
- thunderbolt215 via GitHub
- Freshness
- Last updated February 2026.