metadata

license: apache-2.0
language:
  - en
pipeline_tag: depth-estimation
library_name: coreml
tags:
  - depth
  - relative depth
base_model:
  - depth-anything/Depth-Anything-V2-Small

Depth Anything V2 Small (mlpackage)

In this repo you can find:

The notebook which was used to convert depth-anything/Depth-Anything-V2-Small into a CoreML package.
Both mlpackage files which can be opened in Xcode and used for Preview and development of macOS and iOS Apps
Performence and compute unit mapping report for these models as meassured on an iPhone 16 Pro Max
One model uses internal resolution of 518x518 ("Box") and the other 518x392 ("Landscape").
The "Landscape" is much faster than "Box" but will also give more "juggy" edges, due to the patch I applied to avoid bicubing upsampling (.diff file is also present in this repo)

As a derivative work of Depth-Anything-V2-Small this port is also under apache-2.0

Citation of original work

If you find this project useful, please consider citing:

@article{depth_anything_v2,
  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}
}

@inproceedings{depth_anything_v1,
  title={Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data}, 
  author={Yang, Lihe and Kang, Bingyi and Huang, Zilong and Xu, Xiaogang and Feng, Jiashi and Zhao, Hengshuang},
  booktitle={CVPR},
  year={2024}
}