Knowledge Distillation for Object Detection: from generic to remote sensing datasets
This work applies established techniques to remote sensing data, providing insights for vehicle detection but is incremental as it does not introduce new methods.
The paper evaluated existing knowledge distillation methods for object detection on remote sensing datasets, finding high performance variations and emphasizing the need for result aggregation and cross-validation in this domain.
Knowledge distillation, a well-known model compression technique, is an active research area in both computer vision and remote sensing communities. In this paper, we evaluate in a remote sensing context various off-the-shelf object detection knowledge distillation methods which have been originally developed on generic computer vision datasets such as Pascal VOC. In particular, methods covering both logit mimicking and feature imitation approaches are applied for vehicle detection using the well-known benchmarks such as xView and VEDAI datasets. Extensive experiments are performed to compare the relative performance and interrelationships of the methods. Experimental results show high variations and confirm the importance of result aggregation and cross validation on remote sensing datasets.