Good IOU Score on training data, but bad segmentation on testing data.

[ranjan2601]

I have used inceptionv3 as my backbone and trained it on 20 epochs, got mean iou score of 0.90. However, when i try to predict on test data, the results are like this:

My dataset contains images about tomato disease(10 classes). The Gt mask tells which part on leaf is healthy and which part is diseased, whereas Pr mask is just making the shape of the leaf.
Thanks

[nataliameira]

Hello @ranjan2601,

Could you share how you organized your class Dataset and built your model?

It will be easier to come up with a strategy for your problem.

[tshr-d-dragon]

Hi @ranjan2601,

Please share your training notebook/scripts.

Your input image and its corresponding ground_truth mask do not match. Also, there might be some other issue in the training too. Then, we can provide better solution for you problem statement!!:)