The PyTorch implementation of my graduation project (awarded as Outstanding Graduation Thesis).
The proposed method is inspired by the idea of Slimmable Neural Networks. On the main-stream benchmarks, our tracker (named SlimTrack) consistently performs on-par or better than the baseline tracker SiamRPN++. Note that due to the hierarchical weight sharing design, our parameters are only 40% of SiamRPN++ in the sum of all configurations. Moreover, without any downloading and offloading, the proposed tracker can dynamically switch the computational overhead according to the on-device resource constraints and real-time response requirements.
❤️❤️❤️Our idea is implemented base on the following projects. We really appreciate their wonderful open-source work!