DNN BASED PET IDENTIFICATION FOR HETEROGENEOUS EDGE DEVICES

Our living environment is getting smart every day but most of these solutions focus on humans in the environment. The solution involving pets either indoor or outdoor are very less in the literature. This work tries to fill that gap with the deep neural network-based solution for common pets found in any household like dogs and cats, which could be easily integrated into the existing intelligent devices. Deep neural networks with a knowledge distillation approach have shown significant speed-up and lesser memory consumption where real-time performance is required. The present implementation shows the 20% improvement in terms of accuracy in the same training iterations with the resource constraints.