Design Space Exploration of Low-Bit Quantized Neural Networks for Visual Place Recognition

Visual Place Recognition (VPR) is a critical task for performing global re-localization in visual perception systems, requiring the ability to recognize a previously visited location under variations such as illumination, occlusion, appearance and viewpoint. In the case of robotics, the target devices for deployment are usually embedded systems. Therefore whilst the accuracy of VPR systems is important so too is memory consumption and latency. Recently new works have focused on the Recall@1 metric paying limited attention to resource utilization, resulting in methods that use complex models unsuitable for edge deployment. We hypothesize that these methods can be optimized to satisfy the constraints of low powered embedded systems whilst maintaining high recall performance. Our work studies the impact of compact architectural design in combination with full-precision and mixed-precision post-training quantization on VPR performance. Importantly we not only measure performance via the Recall@1 score but also measure memory consumption and latency. We characterize the design implications on memory, latency and recall scores and provide a number of design recommendations for VPR systems under these limitations.