What is ATSS head in person-detection-0303 model?

[anujonthemove]

All the person-detection-03xx series models mention that they are based on MobileNetV2 backbone with ATSS head. 

I searched for ATSS head for MobileNet and SSD architectures but could not find anything on the internet.

I know that this topic may not be of any relevance as there are so much more capable SOTA object detectors available for use but I am still curious about it.

Can anyone explain or point out to some resource or code or architecture for the topic in question?

[Wan_Intel · Accepted solution]

Hi Anujonthemove,

Thanks for reaching out to us.

For your information, person-detection-0303 is a person detector that is based on MobileNetV2 backbone with an ATSS head for 1280x720 resolution. ATSS here is known as Adaptive Training Sample Selection.

It is a method that allows for an automatic selection of positive and negative samples based on the statistical characteristics of objects. You may refer to the following links for more information. Hope it helps.

1. https://www.scitepress.org/PublishedPapers/2023/117814/117814.pdf
2. https://paperswithcode.com/method/atss#:~:text=Prioritized%20Sampling-,Adaptive%20Training%20Sample%20Selection,-Introduced%20by%20Zhang

Regards,

Wan

[Wan_Intel]

Hi Anujonthemove,

Thanks for your question.

If you need any additional information from Intel, please submit a new question as this thread will no longer be monitored.

Regards,

Wan