I am trying on data set of a footballer, from three different clubs of the same footballer. The goal is to identify the player with his respective clubs. But i am getting a huge error rate. Plz help me tackle this. What next should i do to increase decrease the error rate.
Hi @Aamir Can you be more specific on your dataset? Like, how many images are used to train.
Also, since the error rate is still decreasing try training it further and unfreeze and train even more.
Trainning is around 410, test - around 110
I am not going for many epochs, as @jeremy says showing too much data to our model may over fit the data. So looking for more better understanding…of NN architecture.
Increase in validation loss is a sign of over-fitting. As your loss doesn’t seem to increase. you may try.
Also, what exactly are you trying to predict?
Images of what?
I have images of cristiano ronaldo from three different clubs. Trying to identify the club given what club t shirt he is wearing.
Hi,
So usually the approach for modelling usually is TRY to make a model overfit.
This is a signal that a signal exists and the model is able to learn something.
Once you have achieved overfitting you can use different approaches to reduce that.
However,
In the beginning you should not be worried about overfitting.
- Train it on multiple epochs (Try a range of values 5,10,15,20,25)
- Once the error rate is stopped going down try unfreezing the layers and train the a set of epochs again.
So. In the beginning you WANT the model to over fit. But don’t worry it’s a little difficult to make that happen. Once you’ve tried all of the above test your model out! See if you are having problems with overfitting. In all probability you won’t.