An Anchor-Free Object Detector with Contextual Information for License Plate Detection

Detailed paper site: https://hankerkuo.github.io/Thesis_KSH/chapters/Abstract

What does the model do?

License plate detection & Car pose estimation!

How’s the pose estimation ability?

Assign each pixel to certain class, possible for pixel-wise segmentation!

Results
(a) -> input image	(b) -> High-probability license plate position
(c) -> front pixels	(d) -> rear pixels

What kind of scenario is the model good at?

-> An image with multiple cars inside, see the comparison below

Left column -> Vehicle-detection-based method (Detection missing)

Right column -> Our method (All license plates found)

Model architecture

Backbone

Heads