Awesome
RICNN_RepeatGongCheng-sPaper
This project which contain CNNs of paper is from "Learning Rotation-Invariant Convolutional Neural Networks for Object Detection in VHR Optical Remote Sensing Images", it is peoposed in CVPR 2016, The RICNN extract and learn the rotation-invariant feature.
Usage:
python RICNN.py
And you must set the training dataset path and testing dataset path in RICNN.py at first.
Note:
In here, I set the tensor of input is (227,227,1), so you must reset the model if you want to use color image dataset.
And H5 is used as dataset reading type, its type is (numbers,227,227,channels,number of rotated)
Accuracy of RICNN:
We use rotation-mnist-12k dataset to fed for testing accuracy of RICNN, and the accuracy is 98.03%
Name of dataset:rot-mnist-12K So you should transform the image size in dataset before the run the network.