ultrasound-nerve-segmentation icon indicating copy to clipboard operation
ultrasound-nerve-segmentation copied to clipboard

Accuracy instead of Dice-coeff

Open Rabia-Metis opened this issue 7 years ago • 2 comments

Thanks for the amazing work Can you kindly tell me what will be the customised function for accuracy. I want to use accuracy as a metric instead of dice-coeff

Rabia-Metis avatar Jan 13 '18 13:01 Rabia-Metis

You can define your own loss for accuracy, for example by pixel comparison. I beleive it will be a good solution because you can define the thresholding by yourself

NirDiamant avatar Jan 17 '18 07:01 NirDiamant

Maybe just add metric? model.compile(optimizer=Adadelta(), loss=dice_coef_loss, metrics=[dice_coef,my_custom_metric])

mrgloom avatar Jan 30 '18 07:01 mrgloom