Paper#2 Online Model Distillation for Efficient Video Inference

They proposed online distillation mechanism that is motivated by teacher-student architecture. Student network called JITNet is finetuned with the ground truth that came from Teacher network Mask RCNN.

Online Distillation per n frames. JITNet(Student), Mask RCNN(Teacher)

• Title: Online Model Distillation for Efficient Video Inference
• Paper: https://arxiv.org/abs/1812.02699