|
I am trying to adapt the code provided by Honglak Lee in order to recreate the figures in his ICML 2009 paper (Convolutional deep belief networks for scalable unsupervised learning of hierarchical representations). I have made the changes suggested by Sharath Chandra and xue on this forum (see: this post and this post), but I still can't learn the "parts of faces". I'm starting to wonder if the problem lies with the input parameters, so I was wondering if someone could share their working parameter values? I'm currently using learning rate: 0.01 and number of epochs: 500, and using the Caltech-101 Faces images (resized to 150x130) for training. Any help is greatly appreciated - I'm really stuck with this! |