Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
[CVPR 2025] UniK3D: Universal Camera Monocular 3D Estimation
| Date | Stars |
|---|---|
| 2026-07-24 | 744 |
| 2026-07-25 | 744 |
| 2026-07-28 | 744 |
| 2026-07-30 | 744 |
| 2026-08-06 | 744 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
<div align="center">
# UniK3D: Universal Camera Monocular 3D Estimation
<a href="https://arxiv.org/abs/2503.16591"><img src='https://img.shields.io/badge/arXiv-Paper-red?logo=arxiv&logoColor=white' alt='arXiv'></a>
<a href='https://lpiccinelli-eth.github.io/pub/unik3d'><img src='https://img.shields.io/badge/Project_Page-Website-green?logo=googlechrome&logoColor=white' alt='Project Page'></a>
<a href='https://huggingface.co/spaces/lpiccinelli/UniK3D-demo'><img src='https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Live_Demo-blue'></a>
</div>
<div>
<img src="assets/docs/compose.png" width="100%" alt="Banner 2" align="center">
</div>
<div>
<p></p>
</div>
> [**UniK3D: Universal Camera Monocular 3D Estimation**](https://lpiccinelli-eth.github.io/pub/unik3d),
> Luigi Piccinelli, Christos Sakaridis, Mattia Segu, Yung-Hsu Yang, Siyuan Li, Wim Abbeloos, Luc Van Gool,
> CVPR 2025,
> *Paper at [arXiv 2503.16591](https://arxiv.org/pdf/2503.16591.pdf)*
## News and ToDo
- [ ] Rays to parameters optimization.
- [x] `21.03.2025`: Gradio demo and [Huggingface Demo](https://huggingface.co/spaces/lpiccinelli/UniK3D-demo).
- [x] `20.03.2025`: Training and inference code released.
- [x] `19.03.2025`: Models released.
- [x] `26.02.2025`: UniK3D is accepted at CVPR 2025!
## Visualization
<p align="center">
<img src="assets/docs/intro.gif" alt="animated"/>
</p>
### Single 360 Image
<p align="center">
<img src="assets/docs/venice.gif" alt="animated"/>
</p>
***Check more results in our [website](https://lpiccinelli-eth.github.io/pub/unik3d/)!***
## Installation
Requirements are not in principle hard requirements, but there might be some differences (not tested):
- Linux
- Python 3.10+
- CUDA 11.8+
Install the environment needed to run UniK3D with:
```shell
export VENV_DIR=<YOUR-VENVS-DIR>
export NAME=unik3d
python -m venv $VENV_DIR/$NAME
source $VENV_DIR/$NAME/bin/activate
# Install UniK3D and dependencies (more recent CUDAs work fine)
pip install -e . --extra-index-url https://download.pytorch.org/whl/cu121
# Install Pillow-SIMD (Optional)
pip uninstall pillow
CC="cc -mavx2" pip install -U --force-reinstall pillow-simd
# Install KNN (for evaluation only)
cd ./unik3d/ops/knn;bash compile.sh;cd ../../../
```
If you use conda, you should change the following:
```shell
python -m venv $VENV_DIR/$NAME -> conda create -n $NAME python=3.11
source $VENV_DIR/$NAME/bin/activate -> conda activate $NAME
```
Run UniK3D on the given assets to test your installation (you can check this script as guideline for further usage):
```shell
python ./scripts/demo.py
```
If everything runs correctly, `demo.py` should print: `RMSE on 3D clouds for ScanNet sample: 21.9cm`.
`demo.py` allows you also to save output information, e.g. rays, depth and 3D pointcloud as `.ply` file.
## Gradio Demo
- Plase visit our [HugginFace Space](https://huggingface.co/spaces/lpiccinelli/UniK3D-demo) for an installation-free test on your images!
- You can use a local Gradio demo if the HuggingFace is too slow (CPU-based) by running `python ./gradio_demo.py` after installation.
## Get Started
After installing the dependencies, you can load the pre-trained models easily from [Hugging Face](https://huggingface.co/lpiccinelli) as follows:
```python
from unik3d.models import UniK3D
model = UniK3D.from_pretrained("lpiccinelli/unik3d-vitl") # vitl for ViT-L backbone
```
Then you can generate the metric 3D estimation and rays prediction directly from a single RGB image only as follows:
```python
import numpy as np
from PIL import Image
# Move to CUDA, if any
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = model.to(device)
# Load the RGB image and the normalization will be taken care of by the model
image_path = "./assets/demo/scannet.jpg"
rgb = torch.from_numpy(np.array(Image.open(image_path))).permute(2, 0, 1) # C, H, W
predictions = model.infer(rgb)
# Point Cloud in Camera Coordinate
xyz = predictions["points"]
# UnExcerpt of 9,360 characters
Read on GitHub
5
3
1
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:5bc99719b3c8fa34, topic:computer-vision