I‘am very interested in your perfect work. After scanning it, I found that the part of training the segmentation network with source domain images and annotations is missing.
In the file of train_DGS.py,
Line 64, load_from = "./weights/weights1.h5"
And I tried to train the DGS directly rather than load the weights, I found my results is not so good. The MAE CDR is 0.21, much larger than 0.082. And the Dice optic cup is 0.58 which is much smaller than 0.858. I think maybe it's because I didn't train the segmentation network with source domain images and annotations firstly.
Can you share the code of training the segmentation network or the result of the weights or the weights.h5 file after the training before the train_DGS?
Thanks very much.
I‘am very interested in your perfect work. After scanning it, I found that the part of training the segmentation network with source domain images and annotations is missing. In the file of train_DGS.py, Line 64, load_from = "./weights/weights1.h5"
And I tried to train the DGS directly rather than load the weights, I found my results is not so good. The MAE CDR is 0.21, much larger than 0.082. And the Dice optic cup is 0.58 which is much smaller than 0.858. I think maybe it's because I didn't train the segmentation network with source domain images and annotations firstly. Can you share the code of training the segmentation network or the result of the weights or the weights.h5 file after the training before the train_DGS? Thanks very much.