I’m working with the crnn and a doubt came to me.
How can I have the value of accuracy when predicting an image never seen.
Let me explain better: the network gives me the right text but how do I know how accurately the text is right?
You won’t be able to calculate the accuracy automatically without targets.
Of course you (personally) could have a look at the output and determine, how well your model predicts, but usually you would run a final test of the model using an unseen testset with targets to get an estimation of the “real world performance”.
sorry for late i solve with my softmax function to retrieve the exact value!