CPC G06V 10/7753 (2022.01) [G06V 10/776 (2022.01); G06V 10/7715 (2022.01)] | 23 Claims |
1. A method for training a machine learning model, the method comprising:
training the machine learning model for a first training period with a first training set, the machine learning model trained to receive spaces in images as inputs and to output a plurality of class predictions for the spaces in the images, wherein a space is at least part of one or more images;
identifying a training space having a plurality of regions, wherein a region is a portion of an image;
for each region in the plurality of regions of the training space,
applying the machine learning model to the region to generate one or more class predictions, and
determining a confidence score for the region based on the one or more class predictions for the region, the confidence score indicating a confidence of the machine learning model on the one or more class predictions; and
further training, based on the confidence score for each region in the plurality of regions, the machine learning model for a second training period,
wherein the second training period is after the first training period, and further training the machine learning model comprises:
generating a modified classification loss by modifying a classification loss of each region in the training space based on the confidence score for each region, and
using the modified classification loss to determine one or more values of one or more parameters of the machine learning model.
|