System for updating neural network parameters based on object detection area overlap score

An information processing apparatus comprising:

• at least one processor;
• and
• at least one memory having stored thereon instructions which, when executed by the at least one processor, cause the information processing apparatus at least to:
• acquire data indicating a Ground Truth (GT) area including a position of an object in a target image;
• estimate a plurality of candidate areas where the object is likely to be present in the target image and a first score indicating a possibility of presence of the object in the candidate area by inputting the target image to a learning model for estimating an object area including the position of the object present in the target image;
• calculate a second score indicating the possibility of presence of the object in the candidate area based on a degree of overlap between a candidate area corresponding to a position of interest in the image among the plurality of the estimated candidate areas and an area included in the acquired GT data, and a degree of overlap between another candidate area corresponding to a peripheral area around the position of interest in the image and the acquired data;
• calculate a loss for the learning model based on the second score and the estimated candidate area;
• and
• update a parameter for the learning model based on the calculated loss,
• wherein the learning model is learned by updating the candidate area having the second score greater than or equal to a predetermined value as the GT area in the target image.

An information processing method for learning objectness of an object in an image, the information processing method comprising:

• acquiring an image;
• acquiring Ground Truth (GT) data including at least an object area where an object in the acquired image is present;
• inferring a plurality of candidate areas for the object in the image and a score indicating objectness for the candidate area based on a learning model;
• determining a score indicating objectness for the object area based on a degree of overlap between a candidate area corresponding to a position of interest in the image among the plurality of the inferred candidate areas and an area included in the acquired GT data, and a degree of overlap between another candidate area corresponding to a peripheral area around the position of interest in the image and an area included in the acquired GT data;
• and
• updating a parameter for the learning model based on a loss value calculated based on the determined score and the score indicating objectness for the inferred candidate area.

A non-transitory computer-readable storage medium storing a program for causing a computer to execute an information processing method comprising:

• acquiring an image;
• acquiring Ground Truth (GT) data including at least an object area where an object in the acquired image is present;
• inferring a plurality of candidate areas for the object in the image and a score indicating objectness for the candidate area based on a learning model;
• determining a score indicating objectness for the object area based on a degree of overlap between a candidate area corresponding to a position of interest in the image among the plurality of the inferred candidate areas and an area included in the acquired GT data, and a degree of overlap between another candidate area corresponding to a peripheral area around the position of interest in the image and an area included in the acquired GT data;
• and
• updating a parameter for the learning model based on a loss value calculated based on the determined score and the score indicating objectness for the inferred candidate area.