2021 BRN Discussion, page-10666

Hi ID,

Looks like ARM are doing it in software, and as they use more bits as well it will be slower and use more power than Akida.









I found an ARM patent for compressing images, which seems to be more concerned with restoring the image for viewing rather than, as in Akida's case, classifying the image. Of course, the better the image the more reliable the classification.

US2021126736A1 ENCODING DATA ARRAY

https://worldwide.espacenet.com/patent/search/family/068653488/publication/US2021126736A1?q=US2021126736A1

[0051] In this way, by determining and selecting the (most) appropriate rounding mode from a set of available rounding modes to be applied for a block of data elements, it has been found that it may be possible to facilitate an improved reproduction of the original data, and in particular to reduce the unwanted appearance of coarse steps (or jumps) in the data values that may otherwise occur at least for some encoding schemes, as will be explained further below.