search

alexgkendall / SegNet-Tutorial

Files for a tutorial to train SegNet for road scenes using the CamVid dataset
http://mi.eng.cam.ac.uk/projects/segnet/tutorial.html
847 stars 517 forks source link

Not getting proper output #45

Open chiragmalli opened 8 years ago

chiragmalli commented 8 years ago

Hey, I am training segnet using 74 images that i have taken randomley from camvid dataset.and i have trainied for 3000 iterations with learning rate 0.01. But when i tried to test it is giving me output as i attached the file. Can someone tell me the reason y i am getting this result? figure_1 figure_3

Simon-Prevoteaux commented 7 years ago

Hi,

First off all can you tell me what is the accuracy displayed by caffe during training ?

Have to used the script compute_bn_statistics ?

zhouzhenyou commented 7 years ago

I have met the same issue, what's the reason for this result? I'm quit confused about it.

Simon-Prevoteaux commented 7 years ago

Did you use the compute_bn_statistics script in order to normalize the weights resulting from your training ?

zhouzhenyou commented 7 years ago

Thx for your help. the script has already been used, and the result almost the same to the figure above. My result was almost blue or red full of the figure.

xingbotao commented 7 years ago

I have met the same issue, what's the reason for this result?

jqtao-wh commented 6 years ago

met the same problem... every thing was OK when using python scripts, but using the C++ script 'Scripts/test_segmentation.cpp' got the bad output. Input of C++ script 'Scripts/test_segmentation.cpp' ---> argv[1]model_file: deploy.prototxt (uncommented 'compute_bn_statistics.py' line 189 to line 192) argv[2]trained_file: test_weights.caffemodel (run 'compute_bn_statistics.py') argv[3] img: 0001TP_008550.png argv[4] LUT img: camvid11.png

output_image_

At last, I got the point why 'Scripts/test_segmentation.cpp' output different with 'Scripts/test_segmentation_camvid.py'!

In the proto file 'deploy.prototxt' created by 'Script/compute_bn_statistics.py': --->the last Layer is 'prob' layer, its output is Blob[1,11,height,width], but 'Scripts/test_segmentation.cpp' need ArgMax Layer's output, so I added one ArgMax type layer and got result image same with 'Scripts/test_segmentation_camvid.py' output.