Making an RCNN object detector less picky?

1 view (last 30 days)
Doug
Doug on 10 Jun 2019
Commented: Doug on 19 Jun 2019
I'm attempting to detect objects in images using an RCNN object detector, closely following the stop sign example. The code works, but I need it to be less picky. That is, I want possible matches with lower scores to be included in the output. I can tolerate many more false positives (incorrectly identifying some part of the image as an object) than false negatives (failing to identify an object). Is there a way to specify a lower score threshold for possible matches?
  2 Comments
Shivam Sardana
Shivam Sardana on 18 Jun 2019
Following the stop sign example, after the line,
[bboxes,score,label] = detect(rcnn,testImage,'MiniBatchSize',128)
, you get score of detections. To annotate the data with less detection rate, set value of score according to your requirements.
Doug
Doug on 19 Jun 2019
Every object detected by the "detect" command comes with a bbox and score (the outputs). So far, I'm keeping all of them, regardless of their scores. My question is how to get "detect" to output even more possible objects, with even lower scores. Is there an input that allows me to set the minimum score?

Sign in to comment.

Answers (0)

Products

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!