search

sainatarajan / U-Net

A simple U-Net implementation for custom dataset. Just create required folders and place the images and then start training.
20 stars 6 forks source link

How to run??? #14

Closed kimsangseob closed 4 years ago

kimsangseob commented 4 years ago

Hello, I'm a student and I'm just started about DL. I want to use this code but I don't know how to start.

Can you help me...? Thanks.

kimsangseob commented 4 years ago

Wow....!!!!!!!!!!! Really Thanks!!!! Really...ㅠㅠ

I have final question. Now I used 29 images for training and epoch is 100. Then results are not really good now.

How can I get better results? More training set? More epoch? Or both of them?

sainatarajan commented 4 years ago

You're welcome! There are lots of things to try to improve performance.

  1. Firstly, the data too less. 29 images won't be sufficient. Try to have good quality labeled data since this is supervised learning.
  2. You need to learn about Hyperparameter Optimization. Tweaking some of the parameters for ex: learning rate, epochs, lr decay could give you some improvement.
  3. UNet is a pretty good model. You can also go for other variants of UNet like UNet++.
kimsangseob commented 4 years ago

I don't know how appreciate to you. You are my first teacher about deeplearning.

Big big big biiiiiiiiiiig Thanks to you!