<https://lh3.googleusercontent.com/-jwEXgn6DoGA/V6EbNo80oHI/AAAAAAAAABA/LX7iMBBan9IVuMdG53fCQlOs2ltYXNxJQCLcB/s1600/Krizhevsky%2Bet%2Bal%2B2012.png> I am interested to obtain the top-5 labels and probabilities from the Theano-AlexNet CNN for individual test images. See the above figure from Krizhevsky et al (2012) as an illustrative example how to display such data.
On Monday, August 1, 2016 at 11:53:57 AM UTC-4, AT wrote: > > I have successfully trained and validated Theano-AlexNet, where the CNN > achieves 56.6% validation accuracy for the top-1 class labels and 79.7% for > the top-5, which is in very close agreement with results published in the > arXiv in 2015. > > At this stage I am exploring the python code in an effort > to extract the top-5 class label information for individual test images. > So, far I have been able to run the validate_performance.py program in the > AlexNet folder, i.e., > https://github.com/uoguelph-mlrg/theano_alexnet/blob/master/validate_performance.py, > > and print out the following items: > > 1) filenames for each of the 195 validation hkl image mini-batch files. > > 2) their 256 corresponding validation class labels (i.e., val_labels). > > 3) the single top_5 error value (i.e., error_top_5) for each of the 195 > mini-batches. > > I wonder if you could help me find some additional > information from the CNN: > > 1) the top-5 error rate for each of the 256 images in each of the 195 > mini-batches. > > 2) the for each of the 256 images in each of the 195 mini-batches. > Any comments that you may offer would be greatly > appreciated. > > -- --- You received this message because you are subscribed to the Google Groups "theano-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
