Fetching latest commit…
Cannot retrieve the latest commit at this time.
|Failed to load latest commit information.|
This code works for image denoising without tuning the parameters such as number of convolution layer, learning rate... The clean images should be .png files which save in ./data/real_images, and the noise images save in ./data/dataset/noise_images read_stl10.py reads the binary file of [stl10](https://cs.stanford.edu/~acoates/stl10/) and saves as png files, simultaneously, adds guass noise to each image and saves in ./data/dataset/noise_images utils.py is copy from [https://github.com/carpedm20/DCGAN-tensorflow](https://github.com/carpedm20/DCGAN-tensorflow), which is used to read png files of given directory. dcgan.py is used to train the whole network. python dcgan.py --dataset stl10_binary --batch_size 64 --image_size 96 --epoch 100 --learning_rate 0.00001 --c_dim 3 --dataset, the folder name which you save your images, ./data/stl_binary/real_images. --batch_size, this number is related to final sample images, if you set batch_size is 100, you need change the dcgan.py file line 262 [8, 8] to [10, 10]. --c_dim, which means gray(1) or rgb(3). --image_size --epoch --learning_rate