search

DeepTrial / Retina-VesselNet

A Simple U-net model for Retinal Blood Vessel Segmentation based on tensorflow2
303 stars 76 forks source link

question for training images #10

Closed Kevin-Moon closed 5 years ago

Kevin-Moon commented 6 years ago

hello, thank you very much for sharing the great work.

I wonder why you used 40 images instead of 20 images(originally given for training).

what if we use only 20 images for training?

thank you in advance.

DeepTrial commented 6 years ago

Actually,This model is firstly trained with 20 images,it has almost the same accurary.I use 40 images because

1.i have several test images (non-public database) 2.i want to know if more training images could have higher accuracy.

At last,i found that because of patch-based model,20 more training images only have a very little improvement.but maybe 100 or more can make a difference.