The repository is implementation of our work in "Sparse Low Rank Factorization for Deep Neural Network Compression", Elsevier Neurocomputing, 2020
- Python 3.6 or above
- Tensorflow 1.4.0
- Keras 2.0.1
- Pillow
- Opencv-python
- Cifar10 - Downloaded directly from Keras
- Cifar100- Downloaded directly from Keras
- Dogs vs Cats- Can be downloaded from https://www.microsoft.com/en-us/download/details.aspx?id=54765 or https://www.kaggle.com/c/dogs-vs-cats/data
- MNIST - Downloaded directly from Keras
- Oxford102 - Can be downloaded from http://www.robots.ox.ac.uk/~vgg/data/flowers/102/
- Each folder contain compression codes for SVD, SLR-w, SLR-a, SLR-c
- Datasets for Dogs vs Cats and Oxford102 can be downloaded and kept inside a folder /data/
- Each model can be trained using their corresponding train.py file and further used for compression.
- Parameters rank k, sparsity rate sr and reduction rate rr can be customized within the python files.