This is the implementation of paper "A frequency domain neural network for fast image super-resolution".
File directory:
Network model implememtation:
cnn_fft_train.m This file is for training a new network model.
cnn_init.m The regular-net defined in here.
cnn_fft_test_regular_batch.m This is the testing file for our model.
New layer implementation:
./matlab/+dagnn/poiscale.m Implementation of the regular-net's layers.
Hartely transformation:
hartleyTrans.m Hartely transformation in 2D version.
hartleyTrans3D.m Hartely transformation in 3D version.
Loss function:
vl_nnloss_l3.m
vl_nnloss_sqrt.m
vl_nnloss_expl2.m
vl_nnloss_l2.m
Training data generation:
./generate_training_data/generate_aug_data.m
Testing data:
./testing/BSDS100
./testing/Set5
./testing/Set14
./testing/Set19
Testing Pre-processing (only for testing):
./testing_building_batch/imtobatch.m To divid a testing image into several 360*480 batch. So that it can be process by network.
./testing_building_batch/batchtoim.m To integate many 360*480 batchs into the original image.
Post-processing. Only for testing. (All the state of art in image super-resolution have this post-processing.)
modcrop.m To crop and shave image when testing.
shave.m To crop and shave image when testing.
The below files are used in our Journal version, but not in the arKiv and conference version. File directory:
cnn_init_equalnet.m The equal-net defined in here.
./matlab/+dagnn/HTScale.m Implementation of the equal-net's layers.
Compute the spatial kernel's frequency representation.
HTmatrix.m
computeHT.m
Before runing the code. You need to first finish the compilation of MatConvNet.
Do not re-download a new MatConvNet, since I have modified some files in the library. Directly run: "run matlab/vl_compilenn;" in the matlab will be enough. For more infomation about MatConvNet, please visit here: http://www.vlfeat.org/matconvnet/
MatConvNet is a MATLAB toolbox implementing Convolutional Neural Networks (CNNs) for computer vision applications. It is simple, efficient, and can run and learn state-of-the-art CNNs. Several example CNNs are included to classify and encode images. Please visit the homepage to know more.
In case of compilation issues, please read first the Installation and FAQ section before creating an GitHub issue. For general inquiries regarding network design and training related questions, please use the Discussion forum.