Ocatve Convolution (OctConv)

This is a experiment of OctConv with ResNet-50 on CIFAR-10/CIFAR-100

Y. Chen, H. Fang, B. Xu, Z. Yan, Y. Kalantidis, M. Rohrbach, S. Yan, J. Feng. Drop an Octave: Reducing Spatial Redundancy in Convolutional Neural Networks with Octave Convolution. (2019). https://arxiv.org/abs/1904.05049

Requirment

1. Tensorflow==1.13.1
2. tqdm

Architecture of OctConv

Training

Training network on CIFAR-10/CIFAR-100

python train.py --problem cifar10/cifar100

Training the normal ResNet-50

python train.py --is_octconv=false

Training the OctConv ResNet-50

python train.py --is_octconv=true

Result

Dataset: CIFAR-10

Training loss

Test accuracy

	Parameters
Normal Resnet	38,103,690(100%)
OctConv-Resnet	24,628,068(64.5%)

	alpha	Test accuracy
Normal Resnet	0	84.11%
OctConv-Resnet	0.125	85.36%
OctConv-Resnet	0.25	84.83%
OctConv-Resnet	0.5	82.48%

Reference

1. https://github.com/koshian2/OctConv-TFKeras
2. https://github.com/taki0112/ResNet-Tensorflow