VDIGPKU/DynamicDet

Can it be extended to anchor-free detectors?

Closed this issue · 4 comments

Dear Author:
Hello, first of all, thank you for your excellent work.
You say “This dynamic architecture can be easily adapted to mainstream detectors, e.g., Faster R-CNN and YOLO”, I was wondering if it could be extended to anchor-free detectors such as CornerNet and CenterNet.

Hi, @yangyahu-1994
Theoretically, our approach is orthogonal to anchor-free or anchor-based architectures.
Thus, we believe that it can be extended to anchor-free detectors. However, we have not conducted relevant experiments.

Please feel free to provide feedback if you have any further questions!

OK, thanks very much.
image
In equation (14), $lat_t$ is defined as the target latency, that is, it is the latency of a specific object, such as an autonomous vehicle or a mobile phone, thus allowing for different accuracy-speed trade-offs under different hardware constraints? Can I understand it this way?

Yes!

Thank you very much!