search

vlfeat / matconvnet-fcn

A MatConvNet-based implementation of the Fully-Convolutional Networks for image segmentation
Other
175 stars 125 forks source link

how to fine-tune fcn to my own data #22

Open HLinn opened 8 years ago

HLinn commented 8 years ago

I have 2000 gray pictures. I need to segment the image to two parts,object and background.Now I want to use fcn to do the image segment.How can I fine tune the pretrained fcn models,like pascal-fcn8s-dag.mat to adapt my segment task?

liyin2015 commented 8 years ago

I think you need to load your new .mat, since you are doing semantic segmentation like matconvnet-fcn, change the last several network structure like the file fcnInitializeModel.m did. Actually it is a lot of job. At first get the lmdb data from ur dataset. @HLinn

KirigayaMelon commented 8 years ago

I also want to change the network structure to use for my task,but I couldn't understand the code very clearly.Do I only need to rewrite fcnInitializeModel.m if I want to change structure?@liyin2015

liyin2015 commented 8 years ago

I think so. Like rewrite this sentence, net = vl_simplenn_tidy(load(opts.sourceModelPath)) ; Write a different function to generate the net structure @KirigayaMelon . I found this address will be helpful. http://www.cc.gatech.edu/~hays/compvision/proj6/

KirigayaMelon commented 8 years ago

Thank u.Lots of work to do...@liyin2015

hfarhidzadeh commented 8 years ago

@HLinn @KirigayaMelon Could you tune the network on your data? I have same issue to work on my own data. Could you give me the steps you passed? Thanks

YangFangShu commented 7 years ago

@HLinn @KirigayaMelon I have same issue to work on my own data.Could you tune the network on your data? I am grateful that you can share your code,and my email is :yfs2016@hit.edu.cn .Thanks very much

zhenghaojie666 commented 6 years ago

anyone have solve this problem,Could you give the method. thank you very much!!