Quantcast
Channel: maps for developers - Medium
Viewing all articles
Browse latest Browse all 2230

Double the Resolution vs. Computational Load — Vision SDK

$
0
0

Double the Resolution vs. Computational Load — Vision SDK

By Eric Gundersen

Developers can choose to double the resolution in the new release of the Vision(v0.12.0) — the SDK now provides two different ML models — one optimized for performance on low power chipsets, and one optimized for quality on more expensive embedded systems.

Both models use the same development process, but the “quality” models run at double the resolution (as in 2x the pixels per frame). Since there is a roughly linear relationship between the number of processed pixels and the computational load for a given model, processing images at half the resolution doubles the frames per second (FPS).

This tradeoff is now open for developers to decide. For example, if a developer is targeting devices with a lower-end Qualcomm Snapdragon 650, they can select a performance model to surpass the recommended 5 FPS minimum threshold. A developer working with a more powerful processor, such as the Snapdragon 855 found in the Samsung S10, might utilize the quality model instead. Download iOS/Android and read the full release notes.

The input resolution of the Vision SDK determines the tradeoff between number of images processed per second and the number of pixels processed per image.

eric gundersen (@ericg) | Twitter

Maps feature data from Mapbox and OpenStreetMap and their data partners.


Double the Resolution vs. Computational Load — Vision SDK was originally published in maps for developers on Medium, where people are continuing the conversation by highlighting and responding to this story.


Viewing all articles
Browse latest Browse all 2230

Trending Articles