Sta8is/cityscapes_dav2_depth

Cityscapes leftImg8bit_sequence Depth (Depth Anything V2)

Monocular depth maps for the Cityscapes leftImg8bit_sequence frames, predicted with Depth Anything V2.

Each depth map corresponds 1:1 to a frame in the original Cityscapes leftImg8bit_sequence package (30-frame video snippets around every annotated Cityscapes image, for the train and val splits). This repository contains only the predicted depth maps, not the RGB images — pair them with your own copy of leftImg8bit_sequence from the official Cityscapes download page.

Dataset structure

Depth maps are provided as a single zip archive mirroring the original Cityscapes directory layout:

leftImg8bit_sequence_depthv2.zip
└── leftImg8bit_sequence_depthv2/
    ├── train/
    │   └── <city>/
    │       └── <city>_<seq>_<frame>_leftImg8bit_depth.png
    └── val/
        └── <city>/
            └── <city>_<seq>_<frame>_leftImg8bit_depth.png

• train: 18 cities (aachen, bochum, bremen, cologne, darmstadt, dusseldorf, erfurt, hamburg, hanover, jena, krefeld, monchengladbach, strasbourg, stuttgart, tubingen, ulm, weimar, zurich) — 89,250 images
• val: 3 cities (frankfurt, lindau, munster) — 15,000 images
• Total: 104,250 depth maps (~9.8 GB uncompressed)

File names follow the Cityscapes convention with a _depth suffix appended, e.g.:

aachen_000000_000000_leftImg8bit_depth.png

which corresponds to aachen_000000_000000_leftImg8bit.png in the original leftImg8bit_sequence package.

Image format

• 8-bit single-channel (grayscale) PNG, pixel values in [0, 255] (uint8)
• 2048 × 1024 resolution, matching the original Cityscapes images
• Values are Depth Anything V2's relative (inverse) depth output, normalized to the 8-bit range per image — not metric depth, and not directly comparable in absolute terms across images. Higher pixel values correspond to closer objects. Rescale/normalize as needed for your use case

How this was generated

Depth maps were produced by running the Depth Anything V2 model over every frame of the official Cityscapes leftImg8bit_sequence (train + val) RGB images.

Usage

Download and extract the archive:

huggingface-cli download Sta8is/cityscapes_dav2_depth leftImg8bit_sequence_depthv2.zip --repo-type dataset --local-dir .
unzip leftImg8bit_sequence_depthv2.zip

Used in

This dataset was used in:

• DINO-Foresight: Looking into the Future with DINO
• Advancing Semantic Future Prediction through Multimodal Visual Sequence Transformers

License

This is a derivative product built on top of the Cityscapes Dataset. Use of this data is subject to the Cityscapes Dataset License Agreement — in particular it is restricted to non-commercial purposes (academic research, teaching, personal use). You must independently obtain rights to the original Cityscapes RGB imagery to make use of this data.

Citation

If you use this data, please cite Cityscapes, Depth Anything V2, and the papers that produced this dataset:

@inproceedings{Cordts2016Cityscapes,
  title={The Cityscapes Dataset for Semantic Urban Scene Understanding},
  author={Cordts, Marius and Omran, Mohamed and Ramos, Sebastian and Rehfeld, Timo and Enzweiler, Markus and Benenson, Rodrigo and Franke, Uwe and Roth, Stefan and Schiele, Bernt},
  booktitle={Proc. of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2016}
}

@article{yang2024depth,
  title={Depth Anything V2},
  author={Yang, Lihe and Kang, Bingyi and Huang, Zilong and Zhao, Zhen and Xu, Xiaogang and Feng, Jiashi and Zhao, Hengshuang},
  journal={arXiv:2406.09414},
  year={2024}
}

@article{karypidis2024dinoforesight,
  title={DINO-Foresight: Looking into the Future with DINO},
  author={Karypidis, Efstathios and Kakogeorgiou, Ioannis and Gidaris, Spyros and Komodakis, Nikos},
  journal={arXiv preprint arXiv:2412.11673},
  year={2024}
}

@article{karypidis2025futurist,
  title={Advancing Semantic Future Prediction through Multimodal Visual Sequence Transformers},
  author={Karypidis, Efstathios and Kakogeorgiou, Ioannis and Gidaris, Spyros and Komodakis, Nikos},
  journal={arXiv preprint arXiv:2501.08303},
  year={2025}
}