Image augmentation and object detection

An apparatus for image processing, comprising:

• a processor;
• memory coupled with the processor;
• and
• instructions stored in the memory and executable by the processor to cause the apparatus to:
• manipulate an input image to generate at least one modified input image;
• identify, using an object detection machine learning model, the input image and the at least one modified input image, a first set of objects from the input image and at least one second set of objects corresponding to the at least one modified input image;
• generate, used the object detection machine learning model, a first range of confidence scores corresponding to the first set of objects and at least one second range of confidence scores corresponding to the at least one second set of objects;
• normalize the at least one second range of confidence scores according to the first range of confidence scores;
• identify, from the first range of confidence scores and the normalized at least one second range of confidence scores, a set of confidence scores satisfying a confidence score threshold;
• and
• identify, for each confidence score from the set of confidence scores satisfying the confidence score threshold, the corresponding object as a detected object from the input image.

A non-transitory computer-readable medium storing code for data processing, the code comprising instructions executable by a processor to:

• manipulate an input image to generate at least one modified input image;
• identify, using an object detection machine learning model, the input image and the at least one modified input image, a first set of objects from the input image and at least one second set of objects corresponding to the at least one modified input image;
• generate, used the object detection machine learning model, a first range of confidence scores corresponding to the first set of objects and at least one second range of confidence scores corresponding to the at least one second set of objects;
• normalize the at least one second range of confidence scores according to the first range of confidence scores;
• identify, from the first range of confidence scores and the normalized at least one second range of confidence scores, a set of confidence scores satisfying a confidence score threshold;
• and
• identify, for each confidence score from the set of confidence scores satisfying the confidence score threshold, the corresponding object as a detected object from the input image.