ImageNet Sketch Dataset for Domain Adaptation

ImageNet Sketch Data is a collection of sketch images created by nateraw and hosted on Hugging Face, last updated in May 2022. It contains sketch-style images corresponding to the ImageNet classification challenge categories. The dataset is designed for evaluating model performance on out-of-distribution, non-photographic data.

Use Cases

• Benchmarking classifier accuracy on sketch images versus the original ImageNet photo labels.
• Training domain adaptation models to transfer knowledge from photo features to sketch features.
• Evaluating model robustness to distribution shifts using the sketch-style image data.
• Researching zero-shot or few-shot learning performance across different visual modalities.

Strengths

• Aligned with the established ImageNet taxonomy, providing a structured evaluation framework.
• Hosted on a major platform (Hugging Face) with versioning, last updated in 2022.

Limitations

• Unknown total number of images and sketch samples per class limits reproducibility planning.
• Potential label noise or misalignment if sketch interpretations diverge from original ImageNet photo labels.

Provenance

Source

Hugging Face user nateraw.

Collection Method

Derived sketch images corresponding to ImageNet categories; exact creation method is unknown.

Time Range

null

Freshness

Last updated in May 2022.

Geography

null