Clément Godard

Hi, I agree that this looks suspicious, have you tried swapping them?

Hi, Could you try using a KITTI pretrained model on your gta V data? Usually if the intrinsics are good the model should be able to fine tune on the...

I think that's the one: https://archive.org/details/audio-covers

Hi, Your cameras are not in a stereo configuration I'm afraid. You have to be able to rectify them in order to be able to train with this code

Hi, The camera parameters are not needed for training as the input is assumed to be stereo rectified. How do the predicted depthmaps look like?

> I just wonder that the trained datasets having different parameters with testing image may lead to a wrong result. Maybe I should skip this. It is a known issue,...

How does the training loss look like in tensorboard? the image seems to be very very large

You probably need to modify the dataloader to make it fit how your data is. the main difference in the dataloader between kitti and cityscapes are how filenames are generated...

Looks like you're right! It ends up only adding the horizontal smoothness to the loss. Thanks for spotting this. I'll think of a way to fix this.

@a-jahani @shubhamwagh the simplest way to fix it is to change `range(4)` to `range(8)` in https://github.com/mrharicot/monodepth/blob/master/monodepth_model.py#L355 and https://github.com/mrharicot/monodepth/blob/master/monodepth_model.py#L356 from my experience, it shouldn't really impact the results I will not...