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

TL;DR

ThiNet is a filter-level pruning method for CNNs that efficiently reduces model size and computation without altering network structure, using a novel optimization approach based on next-layer statistics.

Contribution

ThiNet introduces a filter pruning framework based on next-layer statistics, supporting any deep learning library and achieving significant compression and acceleration.

Findings

Achieves 3.31× FLOPs reduction on VGG-16 with minimal accuracy loss.

Reduces over 50% of parameters and FLOPs in ResNet-50 with about 1% accuracy drop.

Prunes VGG-16 to 5.05MB, comparable to AlexNet, with strong 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$\times$ FLOPs reduction and 16.63$\times$ 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.