May 2020 Summaries

Curbside pickup has become an essential aspect of shopping, leading to increased traffic and longer wait times, as retailers struggle to predict customer arrival times, often resulting in lines due to a required 30-minute reservation of pickup spots. Mapbox addresses these challenges by using its Isochrone API to create geofences that alert store staff when customers are nearing, allowing for precise estimates of arrival times and ensuring orders are ready when the customer arrives, enhancing efficiency and reducing customer frustration. Alternative strategies, such as staff monitoring or customer notifications, have proven less effective due to their reliance on human actions, making Mapbox's automated solution a more seamless and reliable approach to managing curbside pickups.

The Mobile AI Compute Engine (MACE) framework, developed by Xiaomi, has been integrated into the Vision SDK v0.12.0 update, expanding support for a wider range of Android devices by including processors from manufacturers like Huawei, Samsung, and Mediatek. This integration enhances device coverage beyond the previously supported Qualcomm Snapdragon Neural Processing Engine (SNPE) to accommodate non-Qualcomm chipsets, thus broadening accessibility for Android users. Qualcomm devices can now select the best inference framework for their needs, while the Vision SDK for iOS continues to support devices as far back as the iPhone 6s, utilizing Apple's CoreML framework.

Mapbox has released version 0.12.0 of its Vision SDK, which offers developers the ability to select between two machine learning models tailored for different devices. One model is optimized for performance on low-power chipsets, while the other is designed for high-quality output on more sophisticated embedded systems. Both models share the same development process, but the quality model operates at double the resolution, resulting in twice the pixels per frame. However, this increased resolution also leads to a higher computational load, aligning with the principle that processing more pixels typically requires more computing power. This update allows developers to make strategic decisions based on the processing capabilities of their target devices, such as opting for performance models to achieve higher frames per second on lower-end chipsets or utilizing quality models for enhanced output on devices with stronger processors.