jtardif999
Regular
Seems to suggest that Renesas currently don’t seem to get or appreciate all that is Akida? They’ve initially bought IP for 2 nodes which limits the scope of its use to simple applications and have obstinately forged ahead with their DRP AI for many use cases that they could also use Akida with more nodes. Eventually market forces will probably dictate the need for them to expand their Akida use cases. This particular acquisition seems to high light a certain arrogance/ignorance on their part. Time will tell. AIMO.this looks like a software classification system run on a CPU/GPU.
US11170215B1 System and method for discriminating and demarcating targets of interest in a physical scene
Reality Analytics
View attachment 8838
View attachment 8839
[0051] A classifier is a set of rules and/or algorithms, executing on a computer processor or other microprocessor-based device or platform, which define how to process data values from samples of a dataset in order to identify the feature represented by said data. A classifier may be “trained” by processing data which is known to represent a particular category of feature and storing the results; these results serve as identifying criteria, and the classifier's rules and algorithms may compare the data of newly processed samples representing an unknown feature with these criteria to determine if the feature is of the same category. In some but not all embodiments, a classifier is trained to identify a specific category of feature and classify an unknown feature as in or out of said category; this category represents the classifier's target of interest.
1. A system for detecting and discriminating a feature of interest merged with other features within a physically transduced scene, the system comprising:
an array generating portion executing on a processor to define a multi-dimensional spatial array containing a plurality of physically transduced samples captured for the scene; and
a target discriminating portion executing on a processor to:
formulate a plurality of classification levels, wherein the plurality of physically transduced samples of the scene are mapped at each of the classification levels into a plurality of unit cells of said classification level, the unit cells of different classification levels encompassing spatial regions of different size within the multi-dimensional spatial array,
apply a plurality of predefined classification schemes at the respective classification levels to generate level-specific classifications for the unit cells thereof, the predefined classification schemes of at least two classification levels being different and mutually independent in execution upon different samples captured from the same portion of the scene,
combine the level-specific classifications for the unit cells across the plurality of classification levels to adaptively construct at least one cluster of spatially-contiguous cluster cells based thereon, the cluster cells each being of a common preselected spatial region size independent of classification level, the at least one cluster at least partially defining the feature of interest in peripheral contour within the scene, and,
trigger a detection signal corresponding to discrimination of the feature of interest within the scene,
wherein the combining of level-specific classifications includes, for each cluster cell within the at least one cluster:
combining the level-specific classifications for the unit cells of each respective classification level contained within the cluster cell to generate a level-specific classification of the cluster cell, and,
combining each level-specific classification of the cluster cell to generate a general classification of the cluster cell.
Akida could do the classification standing on its head.