Tensor Reordering for CNN Compression
This work addresses model compression for CNNs, which is an incremental improvement in reducing computational and memory costs for deployment.
The paper tackles parameter redundancy in Convolutional Neural Networks (CNNs) by pruning in the spectral domain using Discrete Cosine Transform (DCT), achieving significant parameter reduction with minor accuracy loss, as validated on ResNet-50 and MobileNet-V2 for ImageNet classification.
We show how parameter redundancy in Convolutional Neural Network (CNN) filters can be effectively reduced by pruning in spectral domain. Specifically, the representation extracted via Discrete Cosine Transform (DCT) is more conducive for pruning than the original space. By relying on a combination of weight tensor reshaping and reordering we achieve high levels of layer compression with just minor accuracy loss. Our approach is applied to compress pretrained CNNs and we show that minor additional fine-tuning allows our method to recover the original model performance after a significant parameter reduction. We validate our approach on ResNet-50 and MobileNet-V2 architectures for ImageNet classification task.