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

the training graph and "inference1" layer weight #75

Open chanmi opened 8 years ago

chanmi commented 8 years ago

Hello. I'm training the model with fcn8s model (https://github.com/shelhamer/fcn.berkeleyvision.org/tree/master/voc-fcn8s).

I want to check whether my training goes well, so could you show your training graph? especially, my initial loss value is about 3000~4000 but in some iteration the loss is 200000 and like that. and I checked inference1 layer filter's weight, when I see as imagesc in matlab, your model shows diagonal but my initialized model shows diagonal but moved one pixel vertically for all the diagonal pixels. is there anything I have to consider to train?