This repo contains the code for our paper "A novel focal Tversky loss function and improved Attention U-Net for lesion segmentation" accepted at IEEE ISBI 2019.
theta_x is half the (spatial) size of x (due to the strided conv)
g (as well as phi_g) is half the (spatial) size of x (due to max pooling in the unet)
So upsample_g is also half the (spatial) size of x and there is no need to calculate the strides in Conv2DTranspose since it will always be 1, or am I missing something?
The 2x UP step on x_l in Fig. 3 of your paper should be 2x DOWN then, right?
From newmodels.AttnGatingBlock: https://github.com/nabsabraham/focal-tversky-unet/blob/2c6ad9535bf6d1bc73371b0f6a57f7daf0af458e/newmodels.py#L86-L93
theta_x
is half the (spatial) size ofx
(due to the strided conv)g
(as well asphi_g
) is half the (spatial) size ofx
(due to max pooling in the unet)So
upsample_g
is also half the (spatial) size ofx
and there is no need to calculate the strides in Conv2DTranspose since it will always be 1, or am I missing something?The
2x UP
step onx_l
in Fig. 3 of your paper should be2x DOWN
then, right?