Modify aspect ratios in SSD and Retinanet

Do we need to change the aspect ratios according to our dataset in models like SSD and Retinanet?
Similar to what Kmeans does in yolov3?

TLT does not provide the tool to get the exact aspect ratios for customer’s dataset.
You can use the default value. And you can also try to trigger more experiments to monitor the mAP result after changing aspect ratios.
Please Note that you should set the aspect ratios such that the resulting anchor box shapes very roughly correspond to the shapes of the objects you are trying to detect.

1 Like

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.