https://github.com/gaborvecsei/barlow-twins

Clean Tensorflow 2 Implementation of the Barlow Twins self-supervised learning method
https://github.com/gaborvecsei/barlow-twins

embeddings keras python self-supervised-learning semi-supervised-learning tensorflow tensorflow2

Last synced: 27 days ago
JSON representation

Clean Tensorflow 2 Implementation of the Barlow Twins self-supervised learning method

• Host: GitHub
• URL: https://github.com/gaborvecsei/barlow-twins
• Owner: gaborvecsei
• Created: 2021-06-18T16:09:04.000Z (over 3 years ago)
• Default Branch: master
• Last Pushed: 2021-06-28T16:16:38.000Z (over 3 years ago)
• Last Synced: 2024-05-01T13:54:14.444Z (6 months ago)
• Topics: embeddings, keras, python, self-supervised-learning, semi-supervised-learning, tensorflow, tensorflow2
• Language: Jupyter Notebook
• Homepage:
• Size: 3.41 MB
• Stars: 4
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Barlow Twins

Unofficial Tensorflow 2 implementation of the [Barlow Twins Self-Supervised Learning method](https://arxiv.org/abs/2103.03230)

```bash

$ python train.py --name my_test /data

$ python train.py --help

```

```python

model = barlow_twins.BarlowTwinsModel(input_height=224,

                                      input_width=224,

                                      projection_units=8192,

                                      drop_projection_layer=True)

model.load_weights(saved_weights, by_name=True)

# Input image values should be in range [0, 255] --> preprocessing is built into the model

embedding = model(image)

```

# Results

**Convergence (Oxford 102 Flowers**



# Setup

## Pip/Conda

```bash

pip install -r requirements.txt

```

## Docker

**Build**

```bash

docker build -t barlow .

```

**Run a training**

```bash

docker run --rm \

           -t \

           -u $(id -u):$(id -g) \

           --gpus all \

           -v $(pwd):/code \

           -v :/data \

           -w /code \

           barlow \

           python train.py --name my_test /data

```

# Citations

```bibtex

@article{DBLP:journals/corr/abs-2103-03230,

    author    = {Jure Zbontar and Li Jing and Ishan Misra and Yann LeCun and St{\'{e}}phane Deny},

    title     = {Barlow Twins: Self-Supervised Learning via Redundancy Reduction},

    journal   = {CoRR},

    volume    = {abs/2103.03230},

    year      = {2021},

    url       = {https://arxiv.org/abs/2103.03230},

    archivePrefix = {arXiv},

    eprint    = {2103.03230},

    timestamp = {Mon, 15 Mar 2021 17:30:55 +0100},

    biburl    = {https://dblp.org/rec/journals/corr/abs-2103-03230.bib},

    bibsource = {dblp computer science bibliography, https://dblp.org}

}

```

# TODOs

- Evaluation

    - Linear evaluation

    - KNN eval

- Choose or use custom backbone

- Save model

    - Save only when loss improved