Skip to content

Dataset · vsevolodpl

Subjective photo rendering quality benchmark

vsevolodpl/REPID

Large dataset of images and ratings for training models to predict aesthetic rendering quality.

What's inside

The dataset contains between 100 thousand and 1 million records combining images, text, and tabular data. Unlike classic datasets for technical distortions like noise or blur, REPID is designed to model subjective human perception of aesthetics and photo rendering quality.

What you can do with it

  • Training automated aesthetic and image restoration quality assessment (IRQA) models
  • Creating reward models for tuning color correction and photo post-processing algorithms
Tasks
Image classification, Image → image, Other
Size
100K–1M rows
Downloads this month
79.5K
Likes
5

How to load it

from datasets import load_dataset

ds = load_dataset("vsevolodpl/REPID")