This project will no longer be maintained by Intel.
Intel has ceased development and contributions including, but not limited to, maintenance, bug fixes, new releases, or updates, to this project.
Intel no longer accepts patches to this project.
If you have an ongoing need to use this project, are interested in independently developing it, or would like to maintain patches for the open source software community, please create your own fork of this project.
- Usage
- Environment setup
- Sanity checks (Recommended)
- Model files
- Evaluation
- Training
- Gradio demo
- Citation
It is recommended to fetch the latest MiDaS repo via torch hub before proceeding:
import torch
torch.hub.help("intel-isl/MiDaS", "DPT_BEiT_L_384", force_reload=True) # Triggers fresh download of MiDaS repoimport torch
repo = "isl-org/ZoeDepth"
# Zoe_N
model_zoe_n = torch.hub.load(repo, "ZoeD_N", pretrained=True)
# Zoe_K
model_zoe_k = torch.hub.load(repo, "ZoeD_K", pretrained=True)
# Zoe_NK
model_zoe_nk = torch.hub.load(repo, "ZoeD_NK", pretrained=True)Clone this repo:
git clone https://github.com/isl-org/ZoeDepth.git && cd ZoeDepthYou can use local source for torch hub to load the ZoeDepth models, for example:
import torch
# Zoe_N
model_zoe_n = torch.hub.load(".", "ZoeD_N", source="local", pretrained=True)from zoedepth.models.builder import build_model
from zoedepth.utils.config import get_config
# ZoeD_N
conf = get_config("zoedepth", "infer")
model_zoe_n = build_model(conf)
# ZoeD_K
conf = get_config("zoedepth", "infer", config_version="kitti")
model_zoe_k = build_model(conf)
# ZoeD_NK
conf = get_config("zoedepth_nk", "infer")
model_zoe_nk = build_model(conf)##### sample prediction
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
zoe = model_zoe_n.to(DEVICE)
# Local file
from PIL import Image
image = Image.open("/path/to/image.jpg").convert("RGB") # load
depth_numpy = zoe.infer_pil(image) # as numpy
depth_pil = zoe.infer_pil(image, output_type="pil") # as 16-bit PIL Image
depth_tensor = zoe.infer_pil(image, output_type="tensor") # as torch tensor
# Tensor
from zoedepth.utils.misc import pil_to_batched_tensor
X = pil_to_batched_tensor(image).to(DEVICE)
depth_tensor = zoe.infer(X)
# From URL
from zoedepth.utils.misc import get_image_from_url
# Example URL
URL = "https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcS4W8H_Nxk_rs3Vje_zj6mglPOH7bnPhQitBH8WkqjlqQVotdtDEG37BsnGofME3_u6lDk&usqp=CAU"
image = get_image_from_url(URL) # fetch
depth = zoe.infer_pil(image)
# Save raw
from zoedepth.utils.misc import save_raw_16bit
fpath = "/path/to/output.png"
save_raw_16bit(depth, fpath)
# Colorize output
from zoedepth.utils.misc import colorize
colored = colorize(depth)
# save colored output
fpath_colored = "/path/to/output_colored.png"
Image.fromarray(colored).save(fpath_colored)The project depends on :
- pytorch (Main framework)
- timm (Backbone helper for MiDaS)
- pillow, matplotlib, scipy, h5py, opencv (utilities)
Install environment using environment.yml :
Using mamba (fastest):
mamba env create -n zoe --file environment.yml
mamba activate zoeUsing conda :
conda env create -n zoe --file environment.yml
conda activate zoeCheck if models can be loaded:
python sanity_hub.pyTry a demo prediction pipeline:
python sanity.pyThis will save a file pred.png in the root folder, showing RGB and corresponding predicted depth side-by-side.
Models are defined under models/ folder, with models/<model_name>_<version>.py containing model definitions and models/config_<model_name>.json containing configuration.
Single metric head models (Zoe_N and Zoe_K from the paper) have the common definition and are defined under models/zoedepth while as the multi-headed model (Zoe_NK) is defined under models/zoedepth_nk.
Download the required dataset and change the DATASETS_CONFIG dictionary in utils/config.py accordingly.
On NYU-Depth-v2 for example:
For ZoeD_N:
python evaluate.py -m zoedepth -d nyuFor ZoeD_NK:
python evaluate.py -m zoedepth_nk -d nyupython evaluate.py -m zoedepth --pretrained_resource="local::/path/to/local/ckpt.pt" -d nyuPretrained resources are prefixed with url:: to indicate weights should be fetched from a url, or local:: to indicate path is a local file. Refer to models/model_io.py for details.
The dataset name should match the corresponding key in utils.config.DATASETS_CONFIG .
Download training datasets as per instructions given here. Then for training a single head model on NYU-Depth-v2 :
python train_mono.py -m zoedepth --pretrained_resource=""For training the Zoe-NK model:
python train_mix.py -m zoedepth_nk --pretrained_resource=""We provide a UI demo built using gradio. To get started, install UI requirements:
pip install -r ui/ui_requirements.txtThen launch the gradio UI:
python -m ui.appThe UI is also hosted on HuggingFace🤗 here
@misc{https://doi.org/10.48550/arxiv.2302.12288,
doi = {10.48550/ARXIV.2302.12288},
url = {https://arxiv.org/abs/2302.12288},
author = {Bhat, Shariq Farooq and Birkl, Reiner and Wofk, Diana and Wonka, Peter and MĂĽller, Matthias},
keywords = {Computer Vision and Pattern Recognition (cs.CV), FOS: Computer and information sciences, FOS: Computer and information sciences},
title = {ZoeDepth: Zero-shot Transfer by Combining Relative and Metric Depth},
publisher = {arXiv},
year = {2023},
copyright = {arXiv.org perpetual, non-exclusive license}
}