README.md

metadata

license: mit
dataset_info:
  features:
    - name: image
      dtype: image
    - name: commands
      list:
        - name: shape
          dtype: string
        - name: class
          dtype: string
        - name: position_x
          dtype: float64
        - name: position_y
          dtype: float64
        - name: position_z
          dtype: int64
        - name: width
          dtype: float64
        - name: height
          dtype: float64
        - name: color
          sequence: int64
  splits:
    - name: train
      num_bytes: 1838065
      num_examples: 1000
  download_size: 789830
  dataset_size: 1838065
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
language:
  - en
tags:
  - synthetic
size_categories:
  - 1K<n<10K
task_categories:
  - text-to-image
  - text-generation
pretty_name: Houses

Houses

This synthetic dataset contains images of simple houses and their corresponding drawing instructions. It is designed for training models to predict missing elements in visual sequences to facilitate systems such as Copilot.

Features

• Images: 256x256 images representing house drawings.
• Commands: structured commands in JSON format describing the components of the house (e.g., body, roof, windows).

Usage

This dataset can be used to train machine learning models for tasks such as:

• sequence completion;
• image generation from structured commands;
• commands generation from visual inputs.

You can also use code from this repository to generate a dataset of the same format according to your own requirements.