Motion Vector Extrapolation for Video Object Detection

Despite the continued successes of computationally efficient deep neural network architectures for video object detection, performance continually arrives at great trilemma speed versus accuracy computational resources (pick two). Current attempts to exploit temporal information in data overcome this are bottlenecked by state art detection models. This work presents motion vector extrapolation (MOVEX), a technique which performs through use off-the-shelf detectors alongside existing optical flow-based estimation techniques parallel. demonstrates that approach significantly reduces baseline latency any given detector without sacrificing performance. Further reductions up 24 times lower than original can be achieved with minimal loss. MOVEX enables low-latency on common CPU-based systems, thus allowing high-performance beyond domain GPU computing.