search

Went-Liang / UnSniffer

[CVPR 2023] Official Pytorch code for Unknown Sniffer for Object Detection: Don’t Turn a Blind Eye to Unknown Objects
https://arxiv.org/pdf/2303.13769v1.pdf
Apache License 2.0
67 stars 9 forks source link

Questions about the box regression of unknown categories #1

Closed PixelChen24 closed 6 months ago

PixelChen24 commented 1 year ago

I notice that in your paper you don't mention the regression of unknown classes, nor do you explain how to "finetune" the bounding box for high GOC score proposals (in Figure 2).

Do you design your fast r-cnn stage to be class-agnostic so that all proposals share the same box regression parameters?

Looking forward to your answer, thanks a lot.

Went-Liang commented 1 year ago

Thank you for your observation and questions regarding our paper.

While our paper focuses on the detection of unknown objects, we do not specifically address the regression of unknown classes or the fine-tuning of bounding boxes for high GOC score proposals. Our GOC module is class-agnostic, and all proposal boxes share the same box regression parameters. During the training process, the GOC module assigns higher scores to proposals that better enclose objects, indirectly assisting the model in finding more optimal bounding boxes.

Thank you again for your valuable feedback.