How to increase the result accuracy

How can i make changes to improve the accuracy of the captions predicted for images. Should i increase the number of epochs, and can i apply earlystopping. Please if you can suggest me some steps to further increase the accuracy of the predictions to get more accurate captions for the image…?

Hey @saksham_thukral, I don’t think you could increase the accuracy by changing the number of epochs by a large amount. See if you have the computing power, I would say that you could only generate more accurate captions by increasing the vocabulary size. Do you remember sir had set a threshold frequency for building vocab ? I would suggest you to decrease this threshold value to increase the number of words or you can say range of words that our model learns. Only then will our model be a bit more specific regarding the captions for particular images.
NOTE : Your training time would increase massively. Each epoch would take around 40-50 mins while training.

Hope this helps !
Happy Learning :slight_smile:

Also you can experiment with the number of layers and network architecture of the models used.

All the best ! :slight_smile:

i got that , but please if you can clear my one confusion, why in such project , on increasing the number of epochs , the accuracy won’t increase?, i mean if i am increasing the number of epochs, it would result in decreasing the loss, then why it can’t lead to more accurate results, as the loss also gets reduced, So accuracy must increase,…shouldn’t it?

Yes you are right with increase in number of epochs accuracy may increase. But the thing is in this project we do not have any y_test right ? So we cannot measure the accuracy by matching the words. So even if you increase the epochs(decrease the loss more and more), it would only result in rearranging the words in the captions. It would give you better captions no doubt, but if you want more range of words then you do need to increase the vocab size. All the words predicted by your model are directly coming from the vocab you have used. So that’s why I suggested you to try increasing the vocab size. I would suggest you to increase the epochs and see the results. It may result in better rearranging of words and hence better captions :slight_smile:

Keep experimenting !
All the best :slight_smile:

Should i do like->" keep on increasing the number of epochs,till loss is decreasing ",is this a right thougth, or not?

Yes that’s right ! You can keep on iterating till your loss converges !

All the best :slight_smile:

I hope I’ve cleared your doubt. I ask you to please rate your experience here
Your feedback is very important. It helps us improve our platform and hence provide you
the learning experience you deserve.

On the off chance, you still have some questions or not find the answers satisfactory, you may reopen
the doubt.