---
dataset_info:
  features:
  - name: image_1_original
    dtype: image
  - name: image_2_original
    dtype: image
  - name: image_1_inpainted
    dtype: image
  - name: image_2_inpainted
    dtype: image
  - name: image_1_object_mask
    dtype: image
  - name: image_2_object_mask
    dtype: image
  - name: image_1_part_mask
    dtype: image
  - name: image_2_part_mask
    dtype: image
  - name: meta
    dtype: string
  splits:
  - name: test
    num_bytes: 93028739.0
    num_examples: 100
  - name: train
    num_bytes: 4189581042
    num_examples: 5000
  download_size: 4153708060
  dataset_size: 4282609781.0
configs:
- config_name: default
  data_files:
  - split: test
    path: data/test-*
  - split: train
    path: data/train-*
---

The data set for [Mind-the-Glitch: Visual Correspondence for Detecting Inconsistencies in Subject-Driven Generation](https://arxiv.org/abs/2509.21989)