Faster Depth Estimation for Situational Awareness on Urban Streets
Proceedings - 2021 IEEE International Conference on Big Data, Big Data 2021
Depth estimation algorithms are useful components of computer vision systems to assess video streams on urban streets. They can provide important information about the street space and improve situational awareness for humans using the street space. Deep learning algorithms for depth estimation algorithms are slow for these applications. To provide situational awareness to humans, the inference time needs to be small, so that results are fresh and meaningful. This paper explores two promising approaches - pruning and quantization - to improve inference time with a little compromise on the performance. We explore the impact of each of these methods independently and introduce a hybrid method that performs pruning followed by quantization. We evaluate the execution time, resource utilization, and performance of various state-of-the-art depth estimation algorithms with and without one or both of these techniques. We observe that using both pruning and quantization can improve the inference time dramatically, with a 39.6% speedup in inference time and an 81.2% reduction in memory utilization while losing only 4% in performance.
8-bit quantization, depth estimation, edge optimization, layer pruning, smart helmet
Sanjana Srinivas and Mahima Agumbe Suresh. "Faster Depth Estimation for Situational Awareness on Urban Streets" Proceedings - 2021 IEEE International Conference on Big Data, Big Data 2021 (2021): 917-926. https://doi.org/10.1109/BigData52589.2021.9671783