Prediction vs test_gt

[hamrarashiba20]

I am encountering an issue with the prediction versus test_gt (labels border are eliminated) for the 'Sharma' and 'Chen' models in my project. The predictions seem to be inconsistent or incorrect compared to the ground truth labels.

I expect the model to generate a full prediction image that includes the border labels, accurately capturing the entire image.

[eecn]

I haven't seen it for a long time. Maybe it's because the test dataset is only part of the labeled data.