DAE2d

DAE2d

This is the implementation of "Denoising Autoencoder" corresponding to 2-dimensional shape.
Original paper: P. Vincent, H. Larochelle, Y. Bengio, and P.-A. Manzagol. Extracting and composing robust features with denoising autoencoders. In Proceedings of the 25th International Conference on Machine Learning, 2008. link

Usage

1. Build

Please build the source file according to the procedure.

$ mkdir build
$ cd build
$ cmake ..
$ make -j4
$ cd ..

2. Dataset Setting

Recommendation

• Large-scale CelebFaces Attributes (CelebA) Dataset
  This is a large-scale face attributes dataset with more than 200K celebrity images, each with 40 attribute annotations.
  Link: official

Setting

Please create a link for the dataset.
The following hierarchical relationships are recommended.

datasets
|--Dataset1
|    |--train
|    |    |--image1.png
|    |    |--image2.bmp
|    |    |--image3.jpg
|    |
|    |--valid
|    |--test
|
|--Dataset2
|--Dataset3

You should substitute the path of training data for "<training_path>", test data for "<test_path>", respectively.
The following is an example for "celebA".

$ cd datasets
$ mkdir celebA
$ cd celebA
$ ln -s <training_path> ./train
$ ln -s <test_path> ./test
$ cd ../..

3. Training

Setting

Please set the shell for executable file.

$ vi scripts/train.sh

The following is an example of the training phase.
If you want to view specific examples of command line arguments, please view "src/main.cpp" or add "--help" to the argument.

#!/bin/bash

DATA='celebA'

./DAE2d \
    --train true \
    --RVIN true \
    --SPN false \
    --GN false \
    --epochs 300 \
    --dataset ${DATA} \
    --size 256 \
    --batch_size 16 \
    --gpu_id 0 \
    --nc 3

You can select the noise to be added to the input image by switching the "RVIN", "SPN", and "GN".

• Random Valued Impulse Noise (RVIN)
  This noise is generated from random number that follow uniform distribution in the range between the minimum and maximum pixel values.
  In addition, the final pixel value is replaced with the generated pixel value, which the original pixel value does not contribute the final one.

• Salt and Pepper Noise (SPN)
  This noise is generated from random number that follow Bernoulli distribution with the minimum and maximum pixel values.
  In addition, the final pixel value is replaced with the generated pixel value, which the original pixel value does not contribute the final one.

• Gaussian Noise (GN)
  This noise is generated from random number that follow normal distribution with a fixed mean and a fixed standard deviation.
  In addition, the final pixel value is the sum of the generated pixel value and the original pixel value.

Run

Please execute the following to start the program.

$ sh scripts/train.sh

4. Test

Setting

Please set the shell for executable file.

$ vi scripts/test.sh

The following is an example of the test phase.
If you want to view specific examples of command line arguments, please view "src/main.cpp" or add "--help" to the argument.

#!/bin/bash

DATA='celebA'

./DAE2d \
    --test true \
    --RVIN true \
    --SPN false \
    --GN false \
    --dataset ${DATA} \
    --test_in_dir "test" \
    --test_out_dir "test" \
    --size 256 \
    --gpu_id 0 \
    --nc 3

If you want to test the denoising of real image, set "test_in_dir" to "directory of noisy images" and "test_out_dir" to "directory of output ground truth".
In addition, you should turn off the flag of "RVIN", "SPN", and "GN".
However, the two file names must correspond.

Run

Please execute the following to start the program.

$ sh scripts/test.sh