1612.03144.pdf

1. The study focuses on using feature pyramids for object detection.
2. The authors report results on a subset of validation images and a standard test set.
3. The feature pyramid maps an RoI of width w and height h to level Pk using a specific formula.
4. The study uses a weight decay of 0.0001 and a momentum of 0.9.
5. The learning rate is adjusted after a certain number of minibatches.
6. The extra convolutional layers have 256 channel outputs and＊＊＊ AI Research, Cornell University, and Cornell Tech.