Retina-Unet
Retina-Unet copied to clipboard
dismatch of the data_format
Hello: Thanks for your code. When i train as your code, i print the model and found your data_format is channel_first, but maxpooling is channel_last, so your channel dims is dismatch, i correct it and when i training, the loss can be down, but the val_acc always didn't change! Do you know why? i have check the input data and the mask groundtruth like you, and ensure it is no problem.
Hello: Thanks for your code. When i train as your code, i print the model and found your data_format is channel_first, but maxpooling is channel_last, so your channel dims is dismatch, i correct it and when i training, the loss can be down, but the val_acc always didn't change! Do you know why? i have check the input data and the mask groundtruth like you, and ensure it is no problem.
I meet the same problem as you. How many epochs have you run? In my case, the val_acc increases in the 10th epoch. Before the 10th epoch, the val_acc keeps at 0.8968.