Test Image Segmentation

[ectg]

Hi,

I am trying to see the semantic segmentation results for a test image like peppers.png using the pretrained model, pascal-fcn16s-dag.mat. The quickstart guide at http://www.vlfeat.org/matconvnet/quick/ shows how to use DAG for classification, but not segmentation. How would I go about doing this?

Can someone help?

[James1944]

Hi ectg, Were you able to complete this task and get the code running? Do you have any tips? I'm also struggling with getting this working.

Thank you James

[zhangmozhe]

Hi, have you solved this? I am new to matconvnet and just need the segmentation result...

Thanks! Bo

[zisianw]

@ectg @James1944 @zhangmozhe have you guys manages to figure out?

[zhangmozhe]

I found a solution but there's something needed to be changed for different versions of manconvnet. I put the codes below. Also, you can also train the net and modify the fcnTest.m in the matconvnet toolbox.

run matlab/vl_setupnn net = dagnn.DagNN.loadobj(load('pascal-fcn32s-dag.mat')) ; net.mode = 'test' ;

predVar = net.getVarIndex('upscore') ; inputVar = 'data' ; imageNeedsToBeMultiple = false ;
rgb = imread('./matconvnet/car.png') ; im_ = bsxfun(@minus, single(rgb), net.meta.normalization.averageImage) ;

net.evalr({inputVar, im}) ; scores = gather(net.vars(predVar).value) ; [~,pred] = max(scores,[],3) ; imshow(pred,[])

[fanguu]

@zhangmozhe I can't figure out the problem,why I use the code

net= dagnn.DagNN.loadobj(load('pascal-fcn32s-dag.mat')) ; there were following problems 错误使用 dagnn.Layer/load (line 191) No property dilate for a layer of type dagnn.Conv. 出错 dagnn.DagNN.loadobj (line 28) block.load(struct(s.layers(l).block)) ;

[dancivitarese]

Hi @zhangmozhe,

I had the same problem. That's because they didn't update the example after publishing beta23. You should replace your network_init example to a newer version where this parameter is properly given. Em sáb, 8 de out de 2016 às 22:48, fanguu notifications@github.com escreveu:

@zhangmozhe https://github.com/zhangmozhe I can't figure out the problem,why I use the code

net= dagnn.DagNN.loadobj(load('pascal-fcn32s-dag.mat')) ; there were following problems 错误使用 dagnn.Layer/load (line 191) No property dilate for a layer of type dagnn.Conv. 出错 dagnn.DagNN.loadobj (line 28) block.load(struct(s.layers(l).block)) ;

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/vlfeat/matconvnet-fcn/issues/33#issuecomment-252458834, or mute the thread https://github.com/notifications/unsubscribe-auth/ABIcC491CqB0Ic4UM61EUgjy29DAspFoks5qyEf2gaJpZM4I0njs .

Daniel Salles Chevitarese

[zisianw]

@fanguu Try running the code in v1.0beta22, it worked for me. If you want to use it in v1.0beta23, try to follow @dchevitarese 's method.

[mjohn123]

This is my script, it worked in beta 23. I hope it can help for someone who meet the issue above

net = dagnn.DagNN.loadobj(load('data/pascal-fcn32s-dag.mat')) ;
rgbPath='ILSVRC2012_test_00000036.jpeg';
rgb = vl_imreadjpeg({rgbPath}) ;
rgb = rgb{1} ;
% Subtract the mean (color)
im = bsxfun(@minus, single(rgb), net.meta.normalization.averageImage) ;
inputVar = 'data' ;
predVar = net.getVarIndex('upscore') ;
net.eval({inputVar, im}) ;
scores_ = gather(net.vars(predVar).value) ;
[~,pred_] = max(scores_,[],3) ;
displayImage(rgb/255, pred_);

The displayImage function is from fcnTest.m