FCN help needed

[feomat]

Hi all,

i'm quite new to caffe (about 1.5 months now) but i already managed to train a net on my own dataset to classify images. That is quite an easy task. For my master thesis i need a way to segmentate an image.

The easy way which i got working was a pixel classificator but that is not desired because its not precise and really takes a long time to process one image.

So i read so much about segmentation orientated conv nets for weeks now but i could not get one to work. I recently tried the fcn but already fail to understand how the downloaded files from github (https://github.com/shelhamer/fcn.berkeleyvision.org) work. I tried to run the voc-fcn32s folder because how i understood it, i have to train 32s- first then 16s- and after that the 8s- version. I downloaded the caffemodel for this folder which was provided there and the pascal datasets but i dont understand where i have to put the files.

I'm really struggling right now and would really appreciate it if some people could give some insight in how to train this fcn from a newby perspective step by step because there is literally no instruction online how to do this. Maybe something like which files i have to use and where to put because for me its very unclear..

Thanks in advance,

Alex

[Vikraman K]

Hi Alex,

Where you able to train a FCN. If yes, can you please help me. Its urgent. And I'm struggling to find resources. 

Thanks,

K.Vikraman