English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Efficient models for remote sensing object counting are urgently required forapplications in scenarios with limited computing resources, such as drones orembedded systems. A straightforward yet powerful technique to achieve this isknowledge distillation, which steers the learning of student networks byleveraging the experience of already-trained teacher networks. However, itfaces a pair of challenges: Firstly, due to its two-stage training nature, alonger training period is essential, especially as the training samplesincrease. Secondly, despite the proficiency of teacher networks in transmittingassimilated knowledge, they tend to overlook the latent insights gained duringtheir learning process. To address these challenges, we introduce an onlinedistillation learning method for remote sensing object counting. It builds anend-to-end training framework that seamlessly integrates two distinct networksinto a unified one. It comprises a shared shallow module, a teacher branch, anda student branch. The shared module serving as the foundation for both branchesis dedicated to learning some primitive information. The teacher branchutilizes prior knowledge to reduce the difficulty of learning and guides thestudent branch in online learning. In parallel, the student branch achievesparameter reduction and rapid inference capabilities by means of channelreduction. This design empowers the student branch not only to receiveprivileged insights from the teacher branch but also to tap into the latentreservoir of knowledge held by the teacher branch during the learning process.Moreover, we propose a relation-in-relation distillation method that allows thestudent branch to effectively comprehend the evolution of the relationship ofintra-layer teacher features among different inter-layer features. Extensiveexperiments demonstrate the effectiveness of our method.

Remote Sensing Object Counting with Online Knowledge Learning