arXiv: https://arxiv.org/abs/1905.07481
In many application of Compressive Sensing, the signals often posess multi-dimensional structure such as MRI or HSI. Multilinear Compressive Learning is a framework which combines Multidimensional Compressive Sensing and Machine Learning to learn the inference tasks in an end-to-end manner. We demonstrate that MCL is both computationally efficient and accurate, outperforming the vector-based design proposed in Adler et al.
In this repository, we provide the full implementation that was used in our experiments in our work
The following dependencies should be installed prior to running our code:
- tensorflow
- keras
- tqdm
- numpy
The datasets, together with train/validation/test splits can be downloaded from here. After cloning our repository and downloading the data, the data should be put in the directory named "data" at the same level as the "code" directory, i.e.,
MultilinearCompressiveLearningFramework/code MultilinearCompressiveLearningFramework/data
After putting the data to the correct place, run::
bash train.sh
to produce all experiment results