Cityscapes: Urban Street Scenes with Semantic Segmentation and Depth Maps

Cityscapes is a processed version of the original dataset containing 3,475 urban street scene images, converted to HuggingFace datasets format by tanganke. The dataset includes 2,975 training samples and 500 validation samples, with corresponding semantic segmentation labels and depth maps. It was last updated on the platform on September 7, -2025.

Use Cases

• Train semantic segmentation models based on the pixel-level class labels for urban objects.
• Develop depth estimation algorithms based on the provided depth maps.
• Benchmark autonomous driving perception systems using the combined image, segmentation, and depth data.
• Conduct research on multi-task learning for scene understanding using the aligned modalities.

Strengths

• Contains 3,475 total image samples, providing a substantial base for model training.
• Offers a split of 2,975 training and 500 validation images, facilitating standard evaluation workflows.
• Provides multiple aligned data modalities (images, semantic labels, depth maps) for each scene.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality, resolution, and file formats require manual inspection after download.
• The dataset may reflect geographic or urban bias inherent to its original source collection.

Provenance

Source

Processed version of the original Cityscapes dataset, converted by tanganke.

Collection Method

Converted to HuggingFace datasets format; original gathering method is not specified.

Time Range

null

Freshness

Last updated 2025-09-07 16:34:06; freshness should be verified.

Geography

Urban street scenes; specific cities or regions are not detailed.