Workload reduction for non-maximum suppression operation
AssigneeAmazon Technologies, Inc.
InventorsAbinash Mohanty, Randy Renfu Huang
Life of this patent
Estimating the term…
Abstract
A technique for improving the computational time for performing a non-maximum suppression operation may include receiving a request to perform a non-maximum suppression operation on a set of candidate predictions of a computing task, and performing a statistical analysis on a set of confidence scores to reduce the set of candidate predictions. Candidate predictions from the reduced set of candidate predictions are provided as the result of the non-maximum suppression operation.
Claims17
A computing system comprising:
- a host processor coupled to a host memory;
- and
- an acceleration engine having an accelerator circuit and an accelerator memory,
- wherein the computing system is configured to:
- generate a set of confidence scores associated with a set of candidate regions of an image, each of the confidence scores indicating a likelihood that an object of interest is in the corresponding candidate region;
- perform a statistical analysis on the set of confidence scores for the object of interest, the statistical analysis including calculating a mean and a standard deviation of the confidence scores;
- reduce the set of candidate regions into a reduced set of candidate regions by determining a confidence score threshold by adding the mean to a result of multiplying the standard deviation with a user configurable multiplier factor, and discarding candidate regions having a confidence score that is lower than the confidence score threshold to form the reduced set of candidate regions;
- determine which of the candidate regions in the reduced set of candidate regions satisfies an intersection-over-union overlap metric;
- and
- classify the candidate region that satisfies the intersection-over-union overlap metric.
A computer-implemented method comprising:
- receiving a request to perform a non-maximum suppression operation on a set of candidate predictions of a computing task;
- performing a statistical analysis on a set of confidence scores corresponding to the set of candidate predictions to determine a standard deviation of the set of confidence scores;
- determining a confidence score threshold based on the standard deviation;
- discarding candidate predictions having a confidence score below the confidence score threshold to form a reduced set of candidate predictions;
- and
- providing candidate predictions from the reduced set of candidate predictions as a result of the non-maximum suppression operation.
A computer-implemented method comprising:
- receiving a request to perform a non-maximum suppression operation on a set of candidate predictions of a computing task, the candidate predictions associated with respective confidence scores;
- discarding candidate predictions having a confidence score below a confidence score threshold to form a reduced set of candidate predictions;
- discarding additional candidate predictions from the reduced set of candidate predictions based on an intersection-over-union overlap metric;
- and
- providing remaining candidate predictions from the reduced set of candidate predictions as a result of the non-maximum suppression operation.
Dependencies are inferred from the claim text.