---
dataset_info:
  features:
  - name: image
    dtype: image
  - name: detailed_prompt
    dtype: string
  - name: control_seg
    dtype: image
  - name: seg_map
    sequence:
      sequence: uint8
  - name: image_path
    dtype: string
  - name: prompt
    dtype: string
  splits:
  - name: train
    num_bytes: 11095603078.08
    num_examples: 20210
  - name: validation
    num_bytes: 1128604170.0
    num_examples: 2000
  download_size: 7044514076
  dataset_size: 12224207248.08
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
task_categories:
- image-segmentation
---

This dataset is used for training and evaluating conditional image generation models, as described in the paper [Ctrl-U: Robust Conditional Image Generation via Uncertainty-aware Reward Modeling](https://huggingface.co/papers/2410.11236).

Project page: https://grenoble-zhang.github.io/Ctrl-U-Page/