CAE-P: Compressive Autoencoder with Pruning Based on ADMM

Since compressive autoencoder (CAE) was proposed, autoencoder, as a simple and efficient neural network model, has outperformed traditional codecs such as JPEG, JPEG 2000 etc. in lossy image compression. However, it faces the problem that the bitrate, characterizing the compression ratio, cannot be optimized by general methods due to its discreteness. Current research additionally trains an entropy estimator to indirectly optimize the bitrate. In this paper, we proposed the compressive autoencoder with pruning based on ADMM (CAE-P) which replaces the traditionally used entropy estimating technique with ADMM-based pruning method inspired by the field of neural network architecture search and avoided the extra effort needed for training an entropy estimator. Our experiments show that this pruning paradigm helps the CAE-P yield a better result compared with the original CAE along with other traditional codecs when measured in both SSIM and MS-SSIM. We further explored the applied pruning method by looking into the detail of the latent codes learned by CAE-P to examine its effectiveness.