Open ParvaShah opened 5 years ago
@ParvaShah , the Shard files represent the weights of the model. So, if the model is small you should have a smaller number of Shard files. So, I think there is no problem with that I think. If you are getting wrong results when predicting in the browser make sure you are preprocessing the inputs correctly i.e normalization.
I was making my own project on Quick draw Doodle challenge but I'm stuck at a point. When I'm trying to convert my model.h5 file to tensorflow.js compatible file I get only one Shard file and a json file. I have seen many people converting their model files to tensorflowjs compatible file and they end up having 4-5 Shard files. Can you please explain to me why is it so?
I'm getting 97% accuracy on training and 95.08% on testing data but when I load my model to browser its doesn' t perform well or rather performs very poorly. I have just used 3 categories for now.
I guess due to some problem in file conversion my model is performing poorly. Your help would be much appreciated