search

emma-sjwang / BEAL

code for paper Boundary and Entropy-driven Adversarial Learning for Fundus Image Segmentation
MIT License
121 stars 25 forks source link

512 or 800? #7

Closed JACKYLUO1991 closed 4 years ago

JACKYLUO1991 commented 4 years ago

I am very sorry, but there is still a problem that bothers me. The input of the test.py script is 800 and Rescale is not executed. Will this affect the results?

emma-sjwang commented 4 years ago

Our segmentation network is end-to-end and full of convolutional layers. So it could handle different sizes of images.

The random scale is a way of data augmentation and only have 50% chance to perform random scale. The remaining images are not rescaled, so our network can handle the test image without scale with different input size.

JACKYLUO1991 commented 4 years ago

thx...