bdd_train = berkely_driving_dataset(PATH_TO_BERKELY_DATASET, transform=bdd_transforms, type='train', color = False). Why color is false in training script?

fabvio / ld-lsi

Deep learning based lane/freespace detector embedded in ROS node (built for UC3M LSI)

89 stars 18 forks source link

bdd_train = berkely_driving_dataset(PATH_TO_BERKELY_DATASET, transform=bdd_transforms, type='train', color = False). Why color is false in training script? #7

Closed sachinsharma9780 closed 4 years ago

sachinsharma9780 commented 4 years ago

I am confused here, why you have passed color=False in training script. Shouldn't it always be color=True since then only it will take ground truth in which lanes are marked as ego and other lanes. What's the role of black ground truth images?

fabvio commented 4 years ago

Hi, this is intentional. The images loaded during training are not completely black, instead they have contiguous ID for each class (so 0 for background, 1 for ego lane and 2 for other lanes). By the way, this is the way pytorch requires semantic labels.

If you have additional concerns feel free to reopen the issue 👍