Training data format

[hoangtnm]

Hi, when training RainNet, does it need any grounth truth labels (real images), or just 2 pair of composite image and its segmentation mask is enough? thank you

[junleen]

When you are training the network, you should provide the ground truth images as the targets. We didn't use two pairs of composite images and their segmentation masks, just 1 pair.