Edit - The problem is with the open_mask function which is generating correct masks for some images only.
I’m trying to do segmentation on a dataset that is like the cityscape dataset. The masks are given as .png file and it’s being generated for some images but not all.
Is there any way I can get the index to the files being showing by show_batch?
Please point out any possible mistake I might be doing
If you use the Validation dataset, then you can get the index of the files being used by show_batch. In case of training dataset the dataloader randomly shuffles the data, so everytime a new batch is generated. In case of validation set the order of dataset does not change.
To get the name of the files data.valid_ds.items[:32].
Thanks for the help, using a small validation set, I could narrow down the problem. It’s open_mask function that is detecting classes for some images only which doesn’t have the void class, i.e, 255.
np.unique(mask.px.numpy())
>>>array([ 0., 1., 2., 3., 4., 5., 6., 255.], dtype=float32)
np.unique(mask1.px.numpy())
>>>array([0., 1., 2., 3., 4., 5., 6.], dtype=float32)
… seems nice!