https://github.com/fkodom/octconv-pytorch

https://github.com/fkodom/octconv-pytorch

Last synced: about 2 months ago
JSON representation

• Host: GitHub
• URL: https://github.com/fkodom/octconv-pytorch
• Owner: fkodom
• License: gpl-3.0
• Created: 2019-06-12T15:35:25.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2019-06-12T15:36:29.000Z (over 5 years ago)
• Last Synced: 2024-10-23T03:17:17.563Z (2 months ago)
• Language: Python
• Size: 16.6 KB
• Stars: 5
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# OctConv-PyTorch

PyTorch implementation of Octave Convolution layers, as described in: [Drop an Octave: Reducing Spatial Redundancy in Convolutional Neural Networks with Octave Convolution](https://arxiv.org/abs/1904.05049)

### Layers Contained

```

OctConv1d

OctConv2d

OctConv3d

OctConvTranspose1d

OctConvTranspose2d

OctConvTranspose3d

```

Currently, the following keyword arguments for convolution layers are not supported by the implemented OctConv layers:

```

stride

dilation

groups

```

### Example Usage

```python

import torch

from octconv import OctConv2d

batch, chin, nrow, ncol = 10, 16, 64, 64

chout, kernel_size, alpha_in, alpha_out = 32, 3, 0.25, 0.25

x = torch.randn((batch, chin, nrow, ncol))

layer = OctConv2d(chin, chout, kernel_size, alpha_in=alpha_in, alpha_out=alpha_out)

y = layer(x)

```