search

TobyPDE / FRRN

Full Resolution Residual Networks for Semantic Image Segmentation
MIT License
278 stars 93 forks source link

Train own Data #15

Closed JoeHEZHAO closed 7 years ago

JoeHEZHAO commented 7 years ago

Hey, thanks for sharing the code. This is a fantastic idea to do semantic segmentation.

May I ask: is there a way to utilize your code on other data, like my own data ? Is there an instruction for this purpose ?

Best He Zhao

TobyPDE commented 7 years ago

Yes, it is possible. I will soon add some APIs in order to make training on new data easier. Before I do this, you can train on your own data using the newly added train_custom.py file. I added "TODO" instructions where you need to fill in some custom values. The most important thing is to convert your annotation to the correct format. Please see the corresponding remark in the file for further details.

MyVanitar commented 7 years ago

@TobyPDE

Hi, Have you updated the fine-tuning section? We also need a pre-trained model. Also May I ask you how can we fine-tune with different number of classes?

TobyPDE commented 7 years ago

You can fine-tune by replacing the last layer of the network. Initialize all weights to the pre-trained ones and randomly initialize the new last layer. Then train on your own data.