Fast non-maximum suppression algorithm for object detection
A Fast Non-Maximum Suppression (NMS) Algorithm post-processing for object detection includes getting original data output from a deep learning model inference output, the original data including a plurality of bounding boxes, pre-emptively filtering out at least one bounding box of the plurality of bounding boxes from further consideration when applying the algorithm, the at least one bounding box filtered out according to a predetermined criteria, processing data, using sigmoid functions or exponential functions, from bounding boxes of the plurality of bounding boxes not filtered out to generate processed bounding boxes, calculating final scores of the processed bounding boxes, and choosing a processed bounding boxes utilizing the final scores.
1. A method of performing a Fast Non-Maximum Suppression (NMS) Algorithm for object detection comprising:
a processor getting original data output from a deep learning model inference output, the original data including a plurality of bounding boxes;
the processor pre-emptively filtering out at least one bounding box of the plurality of bounding boxes from further consideration when applying the algorithm, the at least one bounding box filtered out according to a predetermined criteria;
after pre-emptively filtering out at least one bounding box of the plurality of bounding boxes from further consideration, the processor processing data, using sigmoid functions or exponential functions, from bounding boxes of the plurality of bounding boxes not filtered out to generate processed bounding boxes;
the processor calculating final scores of the processed bounding boxes; and
the processor choosing a processed bounding boxes utilizing the final scores.
2. The method of claim 1 , further comprising the processor pre-emptively filtering out at least one bounding box of the plurality of bounding boxes from further consideration when the at least one bounding box has a class score according to the deep learning model inference output less than a predetermined first threshold.
3. The method of claim 1 , further comprising the processor pre-emptively filtering out at least one bounding box of the plurality of bounding boxes from further consideration when the at least one bounding box has a class score according to the deep learning model inference output less than a result of applying an inverse_sigmoid function to a predetermined confidence threshold.
4. The method of claim 1 , further comprising the processor pre-emptively filtering out at least one bounding box of the plurality of bounding boxes from further consideration when the at least one bounding box has a class score according to the deep learning model inference output less than a result of applying an inverse_sigmoid function to a product of a predetermined confidence threshold and a predetermined Factor of Scaling.
5. The method of claim 1 , further comprising the processor pre-emptively filtering out at least one bounding box of the plurality of bounding boxes from further consideration when the at least one bounding box, according to the deep learning model inference output, has a product of the result of applying a sigmoid function to the confidence score and the result of applying a sigmoid function to the class score that is less than a predetermined confidence threshold.
6. The method of claim 5 , further comprising the processor pre-emptively filtering out at least one bounding box of the plurality of bounding boxes from further consideration when the at least one bounding box, according to the deep learning model inference output, has a class score that is less than a predetermined second threshold.
7. The method of claim 5 , further comprising the processor determining whether the product of the result of applying a sigmoid function to the confidence score and the result of applying a sigmoid function to the class score is less than a predetermined confidence threshold by using a lookup table indexed according to the confidence score and class score.
8. The method of claim 7 , further comprising the processor pre-emptively filtering out at least one bounding box of the plurality of bounding boxes from further consideration when the at least one bounding box, according to the deep learning model inference output, has a class score that is less than a predetermined third threshold.
9. The method of claim 7 , wherein the predetermined third threshold is calculated by the processor according to a product of a score threshold and a first Factor of Scaling.
10. The method of claim 7 , wherein a class probability for the at least one bounding box depends upon the associated box confidence score, and the predetermined third threshold is calculated by the processor according to a product of the class probability and a second Factor of Scaling.