LightningDataModule for dual CNN architecture


I want to train a double CNN network which consist on a CNN trained with RGB images and another CNN trained with depth images, both from the same Intelrealsense camera.
So, my image folders are like this :

├── depth
│ ├── fail
│ └── success
└── rgb
├── fail
└── success

How to design the train_dataloader method of a LightningDataModule to retrieve both images : one RGB image and its corresponding depth images. The two corresponding images (rgb and depth) have the same name.

And how can I use this LightningDataModule to do such an architecture :

(CNN1 filled with RGB images, CNN2 filled with depth images and concatenation of both CNN to fill a full connected network).