desera/octconv.pytorch

[ICCV 2019] PyTorch implementation of Octave Convolution with pre-trained Oct-ResNet models

octconv.pytorch

PyTorch implementation of Octave Convolution in Drop an Octave: Reducing Spatial Redundancy in Convolutional Neural Networks with Octave Convolution

ResNet-50 on ImageNet

Architecture	LR decay strategy	Parameters	GFLOPs	Top-1 / Top-5 Accuracy (%)
ResNet-50	step (90 epochs)	25.557M	4.089	76.010 / 92.834
ResNet-50	cosine (120 epochs)	25.557M	4.089	77.150 / 93.468
OctResNet-50 (alpha=0.5)	cosine (120 epochs)	25.557M	2.367	77.640 / 93.662

To be Done

• Support for MobileNet family (pending for architectural details from the author)

Acknowledgement

Official MXNet implmentation by @cypw

Citation

@article{chen2019drop,
  title={Drop an Octave: Reducing Spatial Redundancy in Convolutional Neural Networks with Octave Convolution},
  author={Chen, Yunpeng and Fan, Haoqi and Xu, Bing and Yan, Zhicheng and Kalantidis, Yannis and Rohrbach, Marcus and Yan, Shuicheng and Feng, Jiashi},
  journal={arXiv preprint arXiv:1904.05049},
  year={2019}
}