KO | EN

GitLite

Minimalist Open Source Repository Search Engine

gitlite — search

#typescript #ai-agents #developer-tools #cli #codex #claude-code #react #mcp #open-source #deepseek-harness #dsh-plugin #ai

unsupervised-deep-homography

★ 65 Open GitHub ↗

PyTorch implementation of Unsupervised Deep Homography: https://arxiv.org/abs/1709.03966

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Unsupervised Deep Homography - *Unoffical* PyTorch Implementation

[**Unsupervised Deep Homography: A Fast and Robust Homography Estimation
Model**](https://arxiv.org/abs/1709.03966)<br>
Ty Nguyen, Steven W. Chen, Shreyas S. Shivakumar, Camillo J. Taylor, Vijay
Kumar<br>

![models](https://raw.githubusercontent.com/teddykoker/unsupervised-deep-homography/master/figures/models.jpg)
*Figure from original paper. Proposed model is \(c\)*

## Implementation Details

This implementation leverages [kornia](https://github.com/kornia/kornia), an
open source differentiable computer vision library. Kornia is used for computing
the direct linear transform (DLT) as well as the spatial transformation.

Uses [PyTorch Lighting](https://pytorch-lightning.readthedocs.io/en/latest/) for
easy GPU training and reproducibility.

`model.py`: Regression model implementation <br>
`dataset.py`: Synthetic data generator <br>
`train.py`: Train unsupervised model using photometric loss outlined in paper

## Test

Download pre-trained weights
```bash
bash download_weights.sh
```

Create gifs:
```bash
python test.py path/to/test/images
```

Input | Registered
---   | ---
![](https://raw.githubusercontent.com/teddykoker/unsupervised-deep-homography/master/figures/input_0.gif) | ![](https://raw.githubusercontent.com/teddykoker/unsupervised-deep-homography/master/figures/output_0.gif)
![](https://raw.githubusercontent.com/teddykoker/unsupervised-deep-homography/master/figures/input_1.gif) | ![](https://raw.githubusercontent.com/teddykoker/unsupervised-deep-homography/master/figures/output_1.gif)
![](https://raw.githubusercontent.com/teddykoker/unsupervised-deep-homography/master/figures/input_2.gif) | ![](https://raw.githubusercontent.com/teddykoker/unsupervised-deep-homography/master/figures/output_2.gif)
![](https://raw.githubusercontent.com/teddykoker/unsupervised-deep-homography/master/figures/input_3.gif) | ![](https://raw.githubusercontent.com/teddykoker/unsupervised-deep-homography/master/figures/output_3.gif)
![](https://raw.githubusercontent.com/teddykoker/unsupervised-deep-homography/master/figures/input_4.gif) | ![](https://raw.githubusercontent.com/teddykoker/unsupervised-deep-homography/master/figures/output_4.gif)

### Training

**Note: tested on PyTorch version 1.4.0; previous versions have a bug that cause
`torch.inverse() and torch.solve()` to generate runtime errors.**

The model can be trained using synthetic data, created from the
[COCO](http://cocodataset.org/) dataset.

```bash
python train.py path/to/COCO/train/ path/to/COCO/valid/
```

unsupervised-deep-homography

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README