how to integrate object detection models

[Huan80805]

Hi, thanks for providing such great work! I am working a object detection task and I have already trained a object detection model with detectron2 framework. is it possible to integrate detectron2 model into this active learning method ? or more generally, can i integrate object detection model like FasterRCNN yolo?

[Dref360]

Hello,

Thank you for your question.

Object detection is a tricky problem in active learning and hasn't been studied extensively. Niko Sünderhauf made quite a lot of contributions to the domain so we could start there. They have a list of their paper on their blog

We did work on this a few years ago and got good performance using the raw output of SSD and computing the variance between iterations.

This is definitely something that should work using BaaL, so we would be happy to help. We just need someone to lead the initiative. I started something in #188 to make it work with torchvision models.

Happy to chat here, on Slack, or on Zoom!

[pieterblok]

Hi, thanks for providing such great work! I am working a object detection task and I have already trained a object detection model with detectron2 framework. is it possible to integrate detectron2 model into this active learning method ? or more generally, can i integrate object detection model like FasterRCNN yolo?

@Huan80805 I'm actually developing an active learning method for Faster R-CNN in Detectron2 (it's not finished yet, let me know your timeframe, maybe we can work something out).

Actually, the method is based on MaskAL, an active learning framework for Mask R-CNN in detectron2: https://github.com/pieterblok/maskal

MaskAL uses BAAL by the way...