Fully-convolutional image denoiser incorporating camera ISO values as conditional information.
Fully-convolutional image denoiser incorporating camera ISO values as conditional information.
There are two scripts provided: one for runnning a regular fully-convolutional model
and one specific to a GAN architecture.
Running the regular one:
python main.py [path to config file]
If no config file is specified, the default config file in run_configs/default.yaml
is used.
for GAN:
python main_gan.py [path to config file]
The default config file for the GAN architecture is run_configs/default_gan.yaml.
If you want to use your own configuration, modify the config for the relevant
model in a duplicate config file. Explanations of the various parameters are provided in the
comments in the config files.
The dataset comprised of three image classes: buildings, foliage, and text.
Sample denoised images are shown in the table below.
| Noisy | Cleaned |
|---|---|
 |
 |
 |
 |
 |
 |