Abstract

The Probabilistic Object Detection Challenge evaluates object detectionmethods using a new evaluation measure, Probability-based Detection Quality(PDQ), on a new synthetic image dataset. We present our submission to thechallenge, a fine-tuned version of Mask-RCNN with some additionalpost-processing. Our method, submitted under username pammirato, is currentlysecond on the leaderboard with a score of 21.432, while also achieving thehighest spatial quality and average overall quality of detections. We hope thismethod can provide some insight into how detectors designed for mean averageprecision (mAP) evaluation behave under PDQ, as well as a strong baseline forfuture work.

 

Quick Read (beta)