The current deep learning approaches (i.e. CNN) have used huge training data and they shows better performance (such as in PASCAL VOC challenge). So i am a bit confused that do we really need more training data for better performance or do we better model ??...
Thanks
Dear Sanjay Saini
Both of them are needed . If you increase the training data , for sure you will get better performance ( if there is no overfitting)
In the other hand , the model that you will use , will also affect the performance .
In my opinion , an efficient preprocessing and features extraction phases will guide you for the best performance , regardless to the model and how hug data that you are going to use.
Regards,
My suggestion is to increase 'GOOD' training data for that. Good samples are very helpful for training DL networks.
- Badges
- Science topic