RandomHorizontalFlip - masks classes should also be flipped

[dahliau]

Hey,
Thank you very much for this repo. I found it very informative and easy to try!

I noticed an issue in the RandomHorizontalFlip transform.
For the networks which perform classification the order of the lanes does matter (right to left, or left to right).
And the following fix can improve the results for these networks (e.g resa34).

https://github.com/Turoad/lanedet/blob/35665fc9dd8d80df207fa86d2152f9f48618ffe7/lanedet/datasets/process/transforms.py#LL258C1-L259C1

        if 'mask' in sample:
            sample['mask'] = np.fliplr(sample['mask'])
            zeros = sample['mask']==0
            sample['mask'][~zeros] = (1+6)-sample['mask'][~zeros]