Monocular Depth Estimation

Monocular Depth Estimation

Turns a single photo into a depth map using the Savta Depth open source project, with code, data and models kept together. For hobbyists making 3D photos and developers who want to contribute.

Made by kingabzpro on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: Savta Depth is a collaborative open source data science project for estimating depth from a single image. The repository holds the code, the data, the models, the pipelines and the experiments, so you can rebuild the project on another machine and add your own data, models or code. This Space is its Gradio demo.

Who it suits: It suits people who want a depth map from one picture, and developers who want to join an open project on depth estimation.

Industry | Who uses it | How they use it | Example
Photography | Hobbyists | Make a depth map from a photo for 3D photos | Upload a picture and save the depth map
Research | Computer vision students | Study a depth model and its experiments | Open the Colab notebook and run it on a picture
Open source | Contributors | Add data, models or code to the project | Fork the project on DAGsHub and pick a setup route

Skip it if you need video depth or a ready commercial service, because the source describes a single-image project run by contributors.

Where to use it:

• A web browser: go to the Space on Hugging Face and try it with no install
• Google Colab: a notebook loads a model from the project and runs it on a picture you upload
• Windows, macOS or Linux: run a copy with Python, or work in a local environment
• Docker: a development environment built on ML Workspace

Use case | Who | What they get back
Make a depth map | Anyone with a photo | A depth map you can save and download
Make 3D photos | Social media users | A saved map for platforms that support it, such as Facebook
Retrain and test | Contributors | A Colab setup to change the code and run training again

How to set it up: You need a few things in place first.

• Python with pip, on Windows, macOS or Linux, to run a copy of the Space
• An internet connection to fetch the files from Hugging Face
• A DAGsHub account to fork the project, only if you want to contribute

1. Install the Hugging Face hub package.
2. Download the Space into a local folder.
3. Move into that folder.
4. Install Gradio 5.34.0 and the packages listed in requirements.txt.
5. Start the app file.

pip install -U huggingface_hub
hf download kingabzpro/savtadepth --repo-type space --local-dir savtadepth
cd savtadepth
pip install gradio==5.34.0
pip install -r requirements.txt
python app/app_savta.py

To work on the project locally, the guide has you create and activate an environment, install the libraries and pull the data and models with DVC.

make env
source env/bin/activate .
make load_requirements
dvc pull -r origin
dvc checkout

For the Docker route, the guide starts its development environment with a script.

chmod +x run_dev_env.sh
./run_dev_env.sh

How to use it: Online, use the Gradio interface the Space shows. On your own computer, run the app file from the project folder and open the page it serves in your browser. The Colab notebook does the same job on an image you upload.

python app/app_savta.py

Example: You open the Colab notebook, upload a photo and run it. It loads a model from the project and returns the depth map for your picture, which you can save and download.

For newcomers and experts: If this is new to you, start online. Go to the Space, give it one photo and look at the depth map it returns. The guide suggests Colab as the easiest route if you later want to contribute.

If you know the field, note that the project keeps data, models, pipelines and experiments together so a run can be reproduced. It uses DVC to pull data and models, and DAGsHub as the remote. The setup assumes no GPU, so a GPU run may need changed requirements.

You give it one image
The project's model estimates depth
A depth map is returned
You save and download the map
You use it for a 3D photo

Good to know: Google Colab limits how long an instance stays live, which can cut long training runs short.

• License: mit
• Cost to run: none named for the demo, and a DAGsHub account for contributing
• Data it reaches: the image you upload, plus project data and models pulled through DVC from its remote
• Not for: the guide is written for contributors, so the Space is not a full production service

On Hugging Face, 13 November 2026

• Likes: 15
• Built with: gradio
• License: mit

Links

• Space: https://huggingface.co/spaces/kingabzpro/savtadepth
• Files: https://huggingface.co/spaces/kingabzpro/savtadepth/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.