Tinify

Tinify (compress-ay) is a PyTorch library and evaluation platform for end-to-end compression research.

Features

Tinify provides:

• Custom operations, layers and models for deep learning based data compression
• Partial port of the official TensorFlow compression library
• Pre-trained models for learned image compression
• Evaluation scripts to compare learned models against classical image/video compression codecs

Note

Multi-GPU support is currently experimental.

Quick Installation

pip install tinify

For detailed installation instructions, see the Installation Guide.

Getting Started

Encode/Decode Images

Use the provided pre-trained models to compress images:

python examples/codec.py --help

Train Your Own Model

python examples/train.py -d /path/to/dataset/ --epochs 300 -lr 1e-4 --batch-size 16 --cuda --save

See the Training Tutorial for more details.

Evaluate Models

# Evaluate a trained checkpoint
python -m tinify.utils.eval_model checkpoint /path/to/images/ -a $ARCH -p $CHECKPOINT

# Evaluate pre-trained models
python -m tinify.utils.eval_model pretrained /path/to/images/ -a $ARCH -q $QUALITY

Available Models

Model	Paper	Pre-trained
bmshj2018_factorized	Ballé et al. 2018	Yes
bmshj2018_hyperprior	Ballé et al. 2018	Yes
mbt2018_mean	Minnen et al. 2018	Yes
mbt2018	Minnen et al. 2018	Yes
cheng2020_anchor	Cheng et al. 2020	Yes
cheng2020_attn	Cheng et al. 2020	No

See the Model Zoo for the complete list of available models.

Library Structure

tinify/
├── datasets/        # Data loading utilities
├── entropy_models/  # Entropy bottleneck and hyperprior models
├── latent_codecs/   # Latent space encoding/decoding
├── layers/          # Neural network layers (GDN, attention, etc.)
├── losses/          # Rate-distortion loss functions
├── models/          # Compression model architectures
├── ops/             # Custom operations
├── transforms/      # Data transformations
└── zoo/             # Pre-trained model zoo

Citation

If you use Tinify in your research, please cite:

@article{begaint2020tinify,
    title={Tinify: a PyTorch library and evaluation platform for end-to-end compression research},
    author={B{\'e}gaint, Jean and Racap{\'e}, Fabien and Feltman, Simon and Pushparaja, Akshay},
    year={2020},
    journal={arXiv preprint arXiv:2011.03029},
}

License

Tinify is licensed under the BSD 3-Clause Clear License.

Contributing

We welcome contributions! Please read our Contributing Guide before submitting pull requests.