metadata

license: apache-2.0

Depth Anything Core ML Models

See the Files tab for converted models.

Depth Anything model was introduced in the paper Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data by Lihe Yang et al. and first released in this repository.

Online demo is also provided.

Disclaimer: The team releasing Depth Anything did not write a model card for this model so this model card has been written by the Hugging Face team.

Model description

Depth Anything leverages the DPT architecture with a DINOv2 backbone.

The model is trained on ~62 million images, obtaining state-of-the-art results for both relative and absolute depth estimation.

drawing

Depth Anything overview. Taken from the original paper.

Evaluation - Variants

Variant	Parameters	Size (MB)	Weight precision	Act. precision	abs-rel error	abs-rel reference
base-original (PyTorch)	97.5M	390	Float32	Float32
small-original (PyTorch)	24.8M	99.2	Float32	Float32	0.1589	base-original
base-float32	97.5M	194.6	Float32	Float32	0.0056	base-original
base-float16	97.5M	194.6	Float16	Float16	0.0061	base-original
small-float32	24.8M	99.0	Float32	Float32	0.0073	small-original
small-float16	24.8M	45.8	Float16	Float16	0.0077	small-original

Evaluation - Inference time

The following results use the small-float16 variant.

Device	OS	Inference time (ms)	Dominant compute unit
iPhone 14	17.5	160.59	Neural Engine
iPhone 14 Pro Max	17.5	119.33	Neural Engine
iPhone 15	17.0	99.42	Neural Engine
iPhone 15 Pro Max	17.4	116.1	Neural Engine
MacBook Pro (M1 Max)	14.5	32.20	GPU

Download

Install huggingface-hub

pip install huggingface-hub

To download one of the .mlpackage folders to the models directory:

huggingface-cli download \
  --local-dir models --local-dir-use-symlinks False \
  coreml-projects/depth-anything \
  --include "DepthAnythingSmallF16.mlpackage/*"

To download everything, skip the --include argument.