https://github.com/divelab/icnn

https://github.com/divelab/icnn

Last synced: 13 days ago
JSON representation

• Host: GitHub
• URL: https://github.com/divelab/icnn
• Owner: divelab
• Created: 2017-02-11T05:10:55.000Z (over 9 years ago)
• Default Branch: master
• Last Pushed: 2017-04-12T06:08:06.000Z (about 9 years ago)
• Last Synced: 2025-02-26T15:48:56.024Z (over 1 year ago)
• Language: Shell
• Size: 5.1 MB
• Stars: 1
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Rotation and Flip Invariant CNN

Caffe implementation of Rotation and Flip Invariant Convolutional Layer is implemented in caffe_3d @ 9686972.

## Prepare data

[Cremi data](https://cremi.org/data/) is used for training model.

After downloading the data, update the content of train_file and test_file in seg_cremi folder.

## How to run

First define caffe path in train.sh in seg_cremi baseline folder and newmodel folder.

In seg_cremi, there are two folders: baseline and newmodel. You can run train.sh to train the model in case you have installed caffe_3d and prepared the data.

## Kernel Rotation operation:

The kernels in convolutional layers are rotated at eight angles:

![drawing](./assets/KernelRotation.png)

## Kernel Rotation operation:

The kernels in convolutional layers are flipped both horizontally and vertically:

![model](./assets/KernelFlip.png)

## Rotation Invariance with Maxout

![model](./assets/maxoutrotate.png)

## Flip Invariance with Maxout

![model](./assets/maxoutflip.png)

## Model used

![model](./assets/c_model.png)

## Author

[Hongyang Gao](http://eecs.wsu.edu/~hgao/), [Shuiwang Ji](http://www.eecs.wsu.edu/~sji/)