Skip to content

Dataset · saraghznfri

UniSim-Bench dataset collection

saraghznfri/unisim_data

A collection of 15 well-known image quality and similarity datasets combined under the UniSim benchmark.

What's inside

The repository aggregates data for evaluating the UniSim-Bench benchmark (from 100K to 1M records). It includes 15 existing datasets: NIGHTS, BAPPS, PieAPP, ImageReward, HPDv2, AGIQA-3K, MAGICBRUSH, HQ-EDIT, CD-COCO, POLARIS, KADID-10k, KONIQ-10k, PIPAL, SICE, and CIFAR-100-OOO.

What you can do with it

  • Comprehensive evaluation of visual similarity metrics and perceptual image quality
  • Comparing image generation and editing models based on human preferences (HPDv2, ImageReward)
  • Testing reward models for generative neural network alignment
Size
100K–1M rows
Downloads this month
53.7K
Likes
1

How to load it

from datasets import load_dataset

ds = load_dataset("saraghznfri/unisim_data")