Skip to content

Dataset · eugene6923

Diffusion Model Image Understanding Benchmark

eugene6923/Self-Bench

Diagnostic generated image set for evaluating diffusion model classification.

What's inside

The dataset contains 1K to 10K text-image pairs generated by various diffusion models. Built for controlled evaluation of how well diffusion models recognize and understand images from their own domain. Suitable for studying the link between models' generative and discriminative capabilities.

What you can do with it

  • Evaluate diffusion models as zero-shot classifiers
  • Diagnose compositional scene understanding in generative networks
  • Compare recognition quality of own and foreign synthetic images
Size
1K–10K rows
Downloads this month
39.8K
Likes
2

How to load it

from datasets import load_dataset

ds = load_dataset("eugene6923/Self-Bench")