What is your loss function?
Albert X. Wang
1

Hi Albert,

In the Notebook script I load the weights from the vlfeat website :

http://www.vlfeat.org/matconvnet/pretrained/#semantic-segmentation

This way I skip the training as it is heavy to do for such model.

The last section of the notebook compares the results using different settings. Using more deconvolution helps a bit, the vlfeat link also shows a page of different performances.

Show your support

Clapping shows how much you appreciated m.zaradzki’s story.