In object detection, how is the bounding box regressor applied?

Question

Anonymous · Accepted Answer

A regression model is responsible for predicting points. In the case of an
object detection model, this is classically solved with CNN's that predict some
parameterization of a bounding box such as (corner, height, width) in a
supervized manner-- images are paired with bounding box annotations and the
model is restricted to detect a handful of categories. Nowadays, if you want to
do object detection, you want to go the route of DinoV2/3 and CLIP/Molmo, which
are foundation models that have rich representations of images and
vision-langugage mapping making it possible to do open-vocabulary semantic
segmentation -- that's always better than object detection!

In object detection, how is the bounding box regressor applied?

Practice this question with AI

Go Premium

Practice More Questions

Community Answers

Unlock Community Insights