Lesson3-camvid Image segmentation on a custom dateset

I am trying the image segmentation example given at http://localhost:8888/notebooks/course-v3-master/nbs/dl1/lesson3-camvid.ipynb.
I am using a self annotated dataset. But I am getting the following error:
IndexError: Target 33 is out of bounds
I am new to the field and I am not able to understand the cause of the error I am getting.
I have only three classes in the data and the dataset consists of only about 30 images.
Can that be a potential problem?
I will appreciate any help in this regard.
Thank you.


hi @SumukhBansal
How have you labeled your images? are the label classes sequential?(as in 0,1,2, and so on?)
what kind of a dataset are you using? Could you share some information?

Hi Palaash,
I am using PixelAnnotationTool tool to label images. Labels are not sequential.
But, since yesterday I have also tried a binary classification where I have used 0 and 1 as labels, but I was facing the same error. I am using a subset of 11k Hands dataset.

but, are your pixels labels 0,and 1, or 0 and 255 (for binary segmentation)? Fastai expeccts the labels to be sequential. (0,1,2, and so on). That is what creates such an error, usually!

Thank you for pointing that out. It is working now.
Thanks a lot.