UniSim-Bench
UniSim‑Bench is a multimodal perception similarity benchmark created by New York University and EPFL, containing seven multimodal perception similarity tasks across 25 datasets. It covers various image‑to‑text tasks and is designed to evaluate model generalisation across tasks. The benchmark aggregates existing perception tasks and trains models using multi‑task learning. UniSim‑Bench is widely used to assess and improve multimodal perception models, especially for cross‑modal similarity evaluation and generative model quality assessment.
Dataset description and usage context
Dataset Overview
Dataset Name
UniSim‑Bench
Dataset Description
UniSim‑Bench is a comprehensive benchmark covering 7 multimodal perception similarity tasks and 25 datasets. The benchmark is used to evaluate the performance of multimodal perception models and supports model training and evaluation.
Dataset Composition
- Core 2AFC tasks: some datasets are used to train the UniSim model.
- OOD generalisation tasks: all datasets are used only for testing.
Dataset Download
UniSim‑Bench can be found on HuggingFace at this link.
Pair the dataset with AI analysis and content workflows.
Once the source passes your review, move straight into summarization, transformation, report drafting, or presentation generation with the JuheAI toolchain.