Unpaired Facade Images For CycleGAN Translation

Featuring facade images used for unpaired image-to-image translation, originating from the CycleGAN project. It is part of the datasets associated with the 2017 research paper on Cycle-Consistent Adversarial Networks (arXiv:1703.10593). The dataset was uploaded by 'huggan' and last updated in April 2022.

Use Cases

• Train a CycleGAN model for unpaired translation between different facade image styles.
• Apply image-to-image translation techniques using the dataset's facade images as a standard benchmark.
• Use the dataset's text modality for potential multi-modal learning tasks involving facade descriptions.

Strengths

• Dataset is associated with a seminal 2017 research paper (arXiv:1703.10593) on CycleGANs.
• Categorized as a small dataset (Size Category: n<1K) suitable for quick experimentation.
• Includes metadata tags indicating support for multiple data libraries (polars, pandas, datasets).

Limitations

• The dataset is small (Size Category: n<1K), limiting its use for training large-scale models.
• Specific details like row count, column structure, and image resolution are unknown.
• Data is region-tagged as 'us', which may indicate a geographic bias in the facade styles represented.

Provenance

Source

Originally part of the CycleGAN datasets from UC Berkeley.

Collection Method

null

Time Range

null

Freshness

2022-04-12

Geography

Region tag suggests United States ('us'), but specific coverage is unknown.