Depth Pro 2: Sharp and Temporally Consistent Metric Monocular Depth
Paper • 2611.26603 • Published • 26
pip install "git+https://github.com/apple-aiml-research/ml-depth-pro-2"
Load it with any size, for photos and videos without a focal length:
import torch
from huggingface_hub import hf_hub_download
from safetensors.torch import load_file
from depth_pro_2 import DepthPro2
from depth_pro_2.network.fov import FOVEstimator
weights = hf_hub_download("apple/DepthPro2-focal", "model.safetensors")
model = DepthPro2.from_pretrained("apple/DepthPro2-H", precision=torch.bfloat16)
model.fov = FOVEstimator(torch.bfloat16)
model.fov.load_state_dict(load_file(weights))
model = model.to("cuda").eval()
On its own, it estimates the focal length of a photo, or one for the frames of a clip:
from functools import partial
import torch
from huggingface_hub import hf_hub_download
from safetensors.torch import load_file
from depth_pro_2 import load_rgb
from depth_pro_2.model import to_normalized_tensor
from depth_pro_2.network.fov import FOVEstimator
weights = hf_hub_download("apple/DepthPro2-focal", "model.safetensors")
fov = FOVEstimator()
fov.load_state_dict(load_file(weights))
fov = fov.to("cuda").eval()
transform = partial(to_normalized_tensor, device=torch.device("cuda"))
image, _ = load_rgb("photo.jpg")
with torch.no_grad():
f_px = fov.focal_length_px(transform(image)[None]) # pixels of the photo; frames (T, 3, H, W)
The weights are released under the Apple Machine Learning Research Model license, for research use only: non-commercial scientific research and academic development.
@article{DepthPro2:2026:arxiv,
title = {Depth Pro 2: Sharp and Temporally Consistent Metric Monocular Depth},
author = {Peiyun Hu and Marko Mihajlovic and Lyndon Duong and Lahav Lipson and Wei Dong and Ama\"{e}l Delaunoy and Stephan R. Richter and Vladlen Koltun},
journal = {arXiv preprint arXiv:2611.26603},
year = {2026},
url = {https://arxiv.org/abs/2611.26603},
}