We propose GPUSched, a foreground graphics aware preemptive scheduling framework for concurrent DNN inference and seamless graphics rendering on mobile GPUs. Given that existing GPU scheduling methods relying on offline slicing are inadequate for dynamic graphics workloads, GPUSched is driven by two key components: (i) Render-State Detection via GPUPing, which leverages lightweight probing of GPU queue latency to accurately identify foreground rendering operations and (ii) Adaptive Operator Chunking, which dynamically fits DNN model chunk sizes within GPU idle times in a GPU frequency-aware manner. By combining these mechanisms GPUSched, coordinates DNN inference with graphics rendering in real time, preventing deadline violations while maximizing GPU utilization. Experiments on two commodity mobile devices show that GPUSched consistently outperforms baseline schedulers, achieving lower deadline miss rates while sustaining high rendering quality even under demanding graphics workload.

Accurate 3D object detection is crucial for autonomous vehicles and robots to navigate and interact with the environment safely and effectively. Meanwhile, the performance of 3D detector relies on the data size and annotation which is expensive. Consequently, the demand of training with limited labeled data is growing. We explore a novel teacher-student framework employing channel augmentation for 3D semisupervised object detection. The teacher-student SSL typically adopts a weak augmentation and strong augmentation to teacher and student, respectively. In this work, we apply multiple channel augmentations to both networks using the transformation equivariance detector (TED). The TED allows us to explore different combinations of augmentation on point clouds and efficiently aggregates multi-channel transformation equivariance features. In principle, by adopting fixed channel augmentations for the teacher network, the student can train stably on reliable pseudo-labels. Adopting strong channel augmentations can enrich the diversity of data, fostering robustness to transformations and enhancing generalization performance of the student network. We use SOTA hierarchical supervision as a baseline and adapt its dual-threshold to TED, which is called channel IoU consistency. We evaluate our method with KITTI dataset, and achieved a significant performance leap, surpassing SOTA 3D semi-supervised object detection models.

We present a hybrid Visible-Light Communication (VLC) and Radio-Frequency (RF) communication system based on an open-source platform, which is integrated at the Medium-Access-Control (MAC) layer. Downlink data transmissions of the system use VLC with a commercially available lighting source, and uplink transmissions use RF. An auxiliary, low-power infra-red (IR) control channel is also implemented on the uplink to transmit acknowledgement signals for VLC data packets. We experimentally demonstrate that the system has low latency, and can achieve end-to-end TCP/IP throughput comparable to single antenna, 20 MHz 802.11 links when operated in isolation, and provide a 70% increase in total throughput when operating concurrently and in the same channel as a legacy 802.11 TCP/IP link. We also show that the RF signals of our system can achieve fair usage of the RF spectrum with an independent 802.11 network with UDP transmissions. In addition to addressing some practical challenges of implementing hybrid, asymmetric VLC-RF systems, our system is open-source, and requires no custom components and can serve as a reference for researchers who wish to construct and analyze hybrid VLC-RF systems.