Extended Reality (XR) applications demand substantial computational resources, often leading to bulky devices or reliance on external cloud servers. This paper demonstrates the integration of In-Network Computing (INC) into XR applications, specifically for object recognition. By leveraging programmable network devices, such as switches, INC offloads computation from end devices and reduces dependency on cloud servers. This approach enhances the real-time performance of XR systems by minimizing latency, optimizing bandwidth usage, and improving scalability. Our demonstration uses lightweight XR glasses connected to a small-scale network consisting of Raspberry Pi-based switches. The system processes object recognition tasks on both switch and server, showcasing the feasibility and benefits of INC for improving XR experiences while exploring the challenges related to machine learning model optimization and efficient resource utilization. This work contributes to the development of more immersive, responsive, and scalable XR applications in future networked environments.