Click3R

Click3R

Rebuilds a 3D scene from a pair of photos and lets you correct the result with a few clicks that mark the same point in both views. A research demo for computer vision researchers and students.

Made by uva-cv-lab 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: Click3R is a research demo for rebuilding a 3D scene from two photos of it. Pairs of clicks that mark the same point in both pictures steer the model toward a better answer. It sits beside the pretrained DUSt3R model so you can compare the two.

Who it suits: People who work with 3D vision, and students who want to see how a handful of clicks can fix a reconstruction that went wrong.

Industry | Who uses it | How they use it | Example
Research | Computer vision researchers | Compare reconstruction with and without clicks | Open the Cellphone pair and add its one click
Education | Students | See how camera pose errors show up in 3D | Try the Bicycle pair taken from the paper
Hobby 3D work | Curious makers | Test how two photos turn into a 3D view | Load the Toy truck example

People who need a finished scanning app, or a tool for many photos at once, can skip it, because the demo works on pairs.

Where to use it:

• A browser: the Space runs on Hugging Face with nothing to install
• Windows, macOS or Linux: run a copy of the Space with Python 3.12

Use case | Who | What they get back
Test sparse clicks | Researchers | A 3D result that changes as clicks are added
Study failure cases | Students | A side-by-side look at DUSt3R and Click3R
Try a ready pair | Anyone curious | One of 23 example pairs with its clicks

How to set it up:

• Python 3.12 with pip on Windows, macOS or Linux
• The Python packages listed in requirements.txt
• An internet connection on first launch, when the Space fetches its checkpoints

1. Install the Hugging Face hub package.
2. Download the Space's files into a folder.
3. Move into that folder.
4. Install Gradio 6.31.0 and the packages from requirements.txt.
5. Start the app.

pip install -U huggingface_hub
hf download uva-cv-lab/Click3R --repo-type space --local-dir Click3R
cd Click3R
pip install gradio==6.31.0
pip install -r requirements.txt
python app.py

How to use it: Open the Space page in a browser, or start your own copy with the command below. Pick one of the 23 example pairs, which come with their clicks, and compare the 3D result.

python app.py

Example: You choose the Cellphone pair, which uses a single pair of clicks. The README lists a pose error of 3.1 degrees with the click, against 165.0 degrees for DUSt3R on the same pair.

For newcomers and experts: If you are new, open the Space in your browser and try the Cellphone example first. It needs only one click pair, so it is the simplest case to follow.

For experts, the demo pairs the Click3R checkpoint with the DUSt3R ViTLarge comparison weights. Its examples come from the paper figures, human clicks on Hard3DBench pairs, and pairs picked by search where DUSt3R and Click3R without clicks both erred.

You open an image pair
You mark matching points in both views
Click3R uses the clicks to fix the pose
You see the 3D result

Title: Pose error on the Cellphone example in degrees
DUSt3R: 165.0 degrees
Click3R without clicks: 133.5 degrees
Click3R with clicks: 3.1 degrees

Good to know:

• License: cc-by-nc-sa-4, as listed for the Space
• Cost to run: none named
• Data it reaches: checkpoints it downloads from Hugging Face on first launch, plus the example images
• Not for: commercial use of the example images, since their dataset licenses apply

On Hugging Face, 13 November 2026

• Likes: 1
• Built with: gradio
• License: cc-by-nc-sa-4.0

Links

• Space: https://huggingface.co/spaces/uva-cv-lab/Click3R
• Files: https://huggingface.co/spaces/uva-cv-lab/Click3R/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.