… Figure 9 shows the components of a multi-stream, multi-task, video analytics pipeline. The
front… It is typical to include the quantization operation during training. Since quantization is non-…

… Second, we study performance optimizations by exploiting multimodel detection parallelism.
… By applying post-training quantization to use 16 bit half-prevision floating point format (FP16…

… video analytics applications, known as edge video analytics [11]–[14]. For a typical edge …
Jointly optimizing frame scheduling and model caching policy can enhance video analytics …

… for live video analytics since they are designed to optimize human… a video from the spatial
perspective, and quantization … are typical video configuration knobs in video analytics systems […

… The constant quantizer or QPmode is set to 10 for less toleration to the data loss during …
are computationally expensive tasks and can be further optimized as explained in SectionV-B. …

… In experiments, SSDLite with MobileNetV2 backbone without any quantization or model …
, eg, quantization-aware training[44], can be incorporated in the retraining phase to optimize …

… • We propose a novel optimized low-light video analytics system. Through camera encoding
… efficient edge video analytics via frame partitioning and quantization optimization,” in Proc. …

… Lyapunov-based Online Optimization The major difficulty of directly solving P1 is that the …
To address this difficulty, we apply the Lyapunov optimization and define a virtual accuracy …

… Their efficiency gain is quite limited because the optimization is … coefficients that are further
quantized to compress more. … Quantization parameter (QP) is the last parameter to regulate …

… frame size/rate together with the employed DNN (eg, from a pool of DNNs optimized at
different levels with techniques including pruning and quantization), which conducts a tradeoff …