orhan-t / sm-cnn

SM-CNN: Hyperspectral Image Denoising via Self-Modulating Convolutional Neural Networks

Paper

Installation

• Install dependencies:

pip install -r requirements.txt

Train and Test

• Put your data into the 'datasets' directory
• Creat train and val patches:

python get_patch_wdc.py

• Creat test data:

python get_patch_wdc_test.py

• Open the model directory in terminal
• Set python environment
• Train example:

python .\train.py --config-file .\gauss_blind_config.yaml

• Test example:

python .\predict.py --config-file .\gauss_blind_config.yaml

sm-cnn's Issues

Can you provide datasets for me? Thank you.

Fig5

Can you give me Fig 5 relative code to show PSNR and SSIM line graph?
Thank you very much.

Can you tell me how much data I need to prepare for training the model to achieve the score in the paper?

Can you tell me how much data I need to prepare for training the model to achieve the score in the paper?
I tried using the WDC dataset, code automatically processed the data and generated 1024 samples.
When training with these samples, the MPSNR of the model was about 28 or so.
Aftertraining for 50 epochs, when testing with 200 x 200 x 191 size data, the test result is far from the results in the paper, it's only around 21.8. I wonder if it's a problem with too few samples used during training?