Roopa Kottapalli
Aug 26, 2017 · 1 min read

Hi Andrey,

Nice article summarising the different steps to take when encountering problems with NN’s. Was curious if you ever had a situation where the network was training well, the loss was going down across epochs but predictions were totally off.

Am using a densenet to count the number of objects of a particular type(connectors) are there in a photo. The losses do decrease but the final predictions are off by a factor of 10–15.

Any suggestions?

Thanks

Roopa

)

    Roopa Kottapalli

    Written by