Model Training

[cmpsmn]

I am trying to apply your model to slam application. I am using a stereo setup with wide angle camera lenses to obtain the sparse depth samples. With the pretrained models I am getting rather bad results, therefore I am thinking about training the model myself.
I am now questioning if the result will bee good if I use a depth camera for training by taking pictures of an environment similar to the one of the application. Does the result of the training depend on the cameras used or more on the environment? I.e. is it okay to use different cameras for training then in the final application?