PFDet: 2nd Place Solution to Open Images Challenge 2018 Object Detection Track

We present a large-scale object detection system by team PFDet. Our system enables training with huge datasets using 512 GPUs, handles sparsely verified classes, and massive class imbalance. Using our method, we achieved 2nd place in the Google AI Open Images Object Detection Track 2018 on Kaggle.

[elu_18]

Preferred Networks won the 2nd prize at Google AI Open Images contest on Kaggle (with 0.00023 point margin!) and here is how the object detection model is carefully built using Chainer and various techniques. https://t.co/5TrG3gCeqn