Trying to Customize The VOC2012 semantic segmentation RESNET Algorithm to detect solar panels in aerial imagery

Can anyone help answer the question of if it is possible to customisze the semantic segmentstion example on AWS sagemaker to detect solar panels in aerial imagery? I have validation and test dataset and have annotated both also. I was wondering if I can add a new class for the solar panel and do a binary segmentation of solar pixels is = 1 and background = 0 using the sagemaker jupyter notebook example FCN algorthim on sagemaker. I am new to python. I followed the example using my own datat set but got the below error

“CustomerError: Unable to open label file img00000057.png”
" raise CustomerError('Unable to open label file ’ + mask_file_name)"

This error seems to indicate that the path to your image is not correct. You can maybe try to follow this tutorial if you find it easier:

or this one: