TAO SOTA Efficientdet false positives

[user162297]

I’ve trained an efficientdet-d2 model using the TAO toolkit on my custom dataset on a single class, I’m was able to get a 83% mAP while training, the model performed excellent at detecting the object when the object exists with high accuracy, but it also detect other objects that aren’t relevant to the class it was trained on with high accuracy as well.

I’m using the pretrained model efficientdet-b2 as my starting point
learning_rate: 0.003
num_epochs: 80

How can I reduce the false positives?

[Morganh]

@user162297

the model performed excellent at detecting the object when the object exists with high accuracy, but it also detect other objects that aren’t relevant to the class it was trained on with high accuracy as well.

It there an example?

[yingliu]

There is no update from you for a period, assuming this is not an issue anymore.
Hence we are closing this topic. If need further support, please open a new one.
Thanks