How to improve JDetNet SSD network accuracy.

Hi,

Could you please provide more details like, what is the final accuracy you are getting and what is the expectation?

Thanks,
Praveen

Can you pls compare your results with the one listed at the following link?

Hello Soyeb,

The accuracy looks same as TI provided trained model. Please suggest how to improve these accuracy.

Thanks and regards,
Saddam Raut

Dear Soyeb

Thanks for the feedback and the link...

Could you please help us share any recommendations/ techniques / links to improve the accuracy of this model?

What is the accuracy % you are seeing? Is there a place (CDDS?) we can use same video test vectors that you use. So we can compare with TI provided results apple to apple.

We were planning to add create add/ more training to the model. Is there any training data set(s) you'd recommend?

Other than data, we would also like to see if we can optimize the TI provided model. If you could share the design considerations/assumptions for your model, any info on the features / hyper parameters for each / critical layers etc. it will help.

Thirdly wee are planning to enhance our own model and port onto TDA for smart rear camera.

With your inputs, we will be able to identify and tweak a few things to improvise further here.


Kindly advise

Thanks!

Best Regards
Feroz

Thanks a lot Soyeb! Appreciate your inputs.

We have made some progress in the direction and will let you on how it goes. 

Best Regards

Feroz