ThiNet: A Filter Level Pruning Method for Deep Neural Network Compression

TLDR

ThiNet is a unified framework that accelerates and compresses CNNs during training and inference. It performs filter‑level pruning by discarding entire filters deemed unimportant, using statistics from the next layer rather than the current layer, and preserves the original network structure for compatibility with standard libraries. Experiments on ILSVRC‑12 show ThiNet reduces FLOPs by 3.31× and compresses VGG‑16 by 16.63× with only a 0.52% top‑5 accuracy loss, cuts over half the parameters of ResNet‑50 with ~1% loss, and can prune VGG‑16 to a 5.05 MB model that retains AlexNet‑level accuracy while improving generalization.

Abstract

We propose an efficient and unified framework, namely ThiNet, to simultaneously accelerate and compress CNN models in both training and inference stages. We focus on the filter level pruning, i.e., the whole filter would be discarded if it is less important. Our method does not change the original network structure, thus it can be perfectly supported by any off-the-shelf deep learning libraries. We formally establish filter pruning as an optimization problem, and reveal that we need to prune filters based on statistics information computed from its next layer, not the current layer, which differentiates ThiNet from existing methods. Experimental results demonstrate the effectiveness of this strategy, which has advanced the state-of-the-art. We also show the performance of ThiNet on ILSVRC-12 benchmark. ThiNet achieves 3.31 x FLOPs reduction and 16.63× compression on VGG-16, with only 0.52% top-5 accuracy drop. Similar experiments with ResNet-50 reveal that even for a compact network, ThiNet can also reduce more than half of the parameters and FLOPs, at the cost of roughly 1% top-5 accuracy drop. Moreover, the original VGG-16 model can be further pruned into a very small model with only 5.05MB model size, preserving AlexNet level accuracy but showing much stronger generalization ability.