search

torrvision / crfasrnn

This repository contains the source code for the semantic image segmentation method described in the ICCV 2015 paper: Conditional Random Fields as Recurrent Neural Networks. http://crfasrnn.torr.vision/
Other
1.34k stars 462 forks source link

training my own data #123

Open luciaL opened 7 years ago

luciaL commented 7 years ago

I use the crfrnn net to make a segmentation of an organ,so the number of output is 2.Then I modify the example/segmentationcrfrnn/solve.protoxt.Then I run the .sh file.But it check failed at softmax_loss_layer and said number of labels don't match number of predictions.Please help to figure out what's the problem??

Could you tell me how to make 'label' lmdb file?The name of label's size should be 121500500,how to make it?The number of my output is 2,so i need to make it be 12500500,but i don't know how to make it.Thank you very much

bittnt commented 7 years ago

You need to change the name of the deconvolution layer and you also need to change the number of the output.

Make sure you initialize the weights of all layers correctly.

luciaL commented 7 years ago

Could you tell me how to make 'label' lmdb file?The name of label's size should be 121500500,how to make it?The number of my output is 2,so i need to make it be 12500500,but i don't know how to make it.Thank you very much

bittnt commented 7 years ago

https://lmdb.readthedocs.io/en/release/ will probably help you.

lixiang-ucas commented 7 years ago

@bittnt i'm training with my own data, how to initialize the convolution weights?

Sam813 commented 6 years ago

@lixiang-ucas I am also trying to use this network for organ segmentation, could you manage to train the network on your data? so far all my predictions are in black and I don't know whats the proper initialization of the convolutions/deconv?