Multi-task learning for dense object detection

A system, comprising:

• a computing system having one or more processors coupled with memory, configured to:
• identify a biomedical image having at least one feature;
• apply the biomedical image to a feature detection model executed on the one or more processors, the feature detection model comprising a set of parameters to:
• generate a feature map corresponding to the biomedical image,
• generate, using the feature map, a first map identifying a confidence score for at least one pixel as correlated with the at least one feature,
• generate, using an the feature map, a second map identifying an adjustment to the at least one pixel in the first map;
• and
• generate a resultant map based on the first map and a second map, the resultant map including at least one point identifying the at least one feature.

A method, comprising:

• identifying, by a computing system, a training dataset comprising:
• a sample biomedical image having at least one feature,
• a first sample map derived from the sample biomedical image, the first sample map identifying a sample confidence score for at least one sample pixel as correlated with the at least one feature, and
• a second sample map derived from the sample biomedical image, the second sample map identifying a sample adjustment to the at least one sample pixel;
• applying, by the computing system, the sample biomedical image of the training dataset to a feature detection model comprising a set of parameters to:
• generate a feature map corresponding to the sample biomedical image;
• generate, using the feature map, a first map identifying a confidence score for at least one pixel as correlated with the at least one feature,
• generate, using an the feature map, a second map identifying an adjustment to the at least one pixel in the first map;
• and
• determining, by the computing system, an error metric based on the first sample map, the second sample map, the first map, and the second map;
• and
• updating, by the computing system using the error metric, at least one of the set of parameters in the feature detection model.