leggedrobotics/defm
Image Feature Extraction • Updated • 695 • 10
You need to agree to share your Hugging Face information to access this dataset
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
The depth maps are derived from ImageNet images. By requesting access you agree to the ImageNet terms of access (https://www.image-net.org/download.php): use only for non-commercial research and educational purposes, no redistribution. The disparities were predicted with Depth Anything V2 Large, which is licensed CC-BY-NC-4.0.
Log in or Sign Up to review the conditions and access this dataset content.
DeFM benchmark: ImageNet-1k (ILSVRC 2012)
Relative disparity predicted by Depth Anything V2 (ViT-L, input size 518) for every ImageNet-1k train and validation image, truncated to uint16 and stored as lossless 16-bit PNG at the original image resolution (larger = closer). Used for kNN and linear-probe classification.
This is the version used in the benchmarks of DeFM: Learning Foundation Representations from Depth for Robotics (1,281,167 train, 50,000 validation). The original data is from ImageNet-1k (ILSVRC 2012); please cite it (below) together with DeFM.
Use
git clone https://github.com/leggedrobotics/defm && cd defm && pip install -e ".[benchmarks]"
python -m benchmarks.classification --encoder defm_vit_s14
The benchmark downloads this dataset automatically. Class names are in meta.json.
Format
data/<split>-XXXXX-of-XXXXX.parquet:
| column | type | content |
|---|---|---|
id |
string | ImageNet file stem |
label |
class label | class index 0..999 (sorted WordNet ids, see meta.json) |
depth |
image | 16-bit PNG, uint16 relative disparity |
Citation
@misc{patel2026defm,
title={DeFM: Learning Foundation Representations from Depth for Robotics},
author={Manthan Patel and Jonas Frey and Mayank Mittal and Fan Yang and Alexander Hansson and Amir Bar and Cesar Cadena and Marco Hutter},
year={2026}, eprint={2601.18923}, archivePrefix={arXiv}, primaryClass={cs.RO}, url={https://arxiv.org/abs/2601.18923}
}
@inproceedings{deng2009imagenet, title={ImageNet: A Large-Scale Hierarchical Image Database}, author={Deng, Jia and Dong, Wei and Socher, Richard and Li, Li-Jia and Li, Kai and Fei-Fei, Li}, booktitle={CVPR}, year={2009}}
@inproceedings{yang2024depthv2, 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}, booktitle={NeurIPS}, year={2024}}
- Downloads last month
- 19
Models trained or fine-tuned on leggedrobotics/defm-bench-imagenet-depth