Transfer Learning Algorithm for Object Detection

Yuta Suzuki, Daiki Kuyoshi, Satoshi Yamane


This study is related to transfer learning in Faster-RCNN, which is a representative model for object detection tasks. Image recognition includes image classification, object detection and image segmentation task. Transfer learning is especially important for the object detection task and the image segmentation task because of the high cost of generating training data. In this study, we use an algorithm to calculate the difference between tasks by focusing on the amount of parameter updates. We then applied the algorithm to an object detection task and aimed to make it useful for transfer learning.

Full Text:



  • There are currently no refbacks.