2nd Place Solution in Google AI Open Images Object Detection Track 2019
This work addresses object detection in large-scale datasets for computer vision practitioners, but it is incremental as it builds on existing methods with minor innovations.
The authors tackled the Google AI Open Images object detection challenge by combining multiple strategies like multi-scale training and FPN with a ResNet200-vd backbone, achieving a public leaderboard score of 0.6269 with single-scale testing and securing 2nd place through a novel top-k voting-NMS method for merging model results.
We present an object detection framework based on PaddlePaddle. We put all the strategies together (multi-scale training, FPN, Cascade, Dcnv2, Non-local, libra loss) based on ResNet200-vd backbone. Our model score on public leaderboard comes to 0.6269 with single scale test. We proposed a new voting method called top-k voting-nms, based on the SoftNMS detection results. The voting method helps us merge all the models' results more easily and achieve 2nd place in the Google AI Open Images Object Detection Track 2019.