How can I improve this model for better results? Or what can I do generally to ensure better results with maximum accuracy?
Improvement of the model
Why you guys don’t reply to the doubts? I have been asking from a long time and you haven’t responded to my queries…neither this one nor the previous ones.
Sorry for this @rishitagoel13.
Before i let you know the way to improve model performance , can you just brief me once about the model that you are currently using.
This the model that they taught in the videos but it is not showing that much accuracy. I am taking the 15th epoch weight. I wanna improve it so pls suggest me how to do it. Do I need to make changes to this model or overall some other changes or dataset?
Can you please reply fast I have to complete my project.
hey @rishitagoel13 ,
sorry for late replies , due to diwali i was on leave so couldn’t check your doubts.
Can you please share me a screenshot of your current training metrics also.
your trainings results.
hey @rishitagoel13,
You can work a bit more on model , like adding more layers , increasing its complexity and might be instead of using a pre trained resnet model , you can train it yourself and used the customized one. Or may be instead of using ResNet you can try other pre trained models too.
But i am not sure that using other models will help in improving performance ,you can give that a try.
Mainly i think to improve the overall performance , you can try using more bigger dataset. That would surely be helpful.
Other then these there are a lot of other changes that can be made too.
- Changing/customizing optimizer
- Instead of pre trained embeddings , train your own .
- No. of epochs
- Applying cross validation
- Changing batch sizes
and lot more…
So give them a try and see how it works out. It might decrease the performance too. So always keep backup files.
What other datasets I can use instead of flickr?
there two flickr datasets ,
one a small subset and another is the full complete dataset.
So , you can try with that one.