Among the state-of-the-art neural network pruning methods, we looked at the papers that approached weights in convolutional layers from a new value of "geometric median" rather than "small-norm-less-important".
He, Yang, et al. "Filter pruning via geometric median for deep convolutional neural networks acceleration." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019. Filter pruning via geometric median
Please visit my co-researcher's github as well! (he's~ hansome)
https://github.com/cloudpark93
- Python 3.8
- Keras 2.4.3
- kerassurgeon 0.2.0
VGG16-cifar10
This is a customized model that can train CIFAR10, not the ImageNet dataset, and we implemented it based on easy-filter-pruning.
Original ResNet
We proceeded with the pre-trained model provided by Keras ResNet.
I will add tests for models and datasets from ResNet56(CIFAR10) and ResNet18(ImageNet).