zhongetal
Final policy RL fine-tuning Envir
Figure dynamics into predict learning from green.
20885c72ca35d75619d6a378edea9f76-Paper.pdf
Object detection has achieved promising success, but requires large-scale fullyannotated data, which is time-consuming and labor-extensive. Therefore, we consider object detection with mixedsupervision, which learns novelobject categories using weak annotations with thehelpoffullannotations ofexistingbase objectcategories.