Methods and apparatus for multi-task recognition using neural networks

A method to perform multi-task recognition comprising:

• building, by executing an instruction with at least one processor, layers of a neural network to generate a facial identifier feature map based on image data, the facial identifier feature map to identify facial features of a face within the image data, the building of the layers including:
• reducing redundant filters corresponding to lower layers of the neural network;
• learning a residual function reference to a layer input;
• and
• broaden a network width and provide a multi-scale feature extraction property to increase emotion recognition;
• generating, by executing an instruction with the at least one processor, attribute feature maps based on the facial identifier feature map, the attribute feature maps associated with facial attributes;
• calculating, by executing an instruction with the at least one processor, probability factors for respective ones of the facial attributes of the attribute feature map, the probability factors identifying the respective facial attributes;
• classifying, by executing an instruction with the at least one processor, an emotion based on the probability factors for the respective facial attributes;
• and
• generating, by executing an instruction with the at least one processor, a report indicative of the classified emotion.

A non-transitory computer readable storage medium comprising instructions which, when executed, cause a machine to at least:

• build layers of a neural network to generate a facial identifier feature map based on image data, the facial identifier feature map to identify facial features of a face within the image data, the building of the layers including:
• reducing redundant filters corresponding to lower layers of the neural network;
• learning a residual function reference to a layer input;
• and
• broaden a network width and provide a multi-scale feature extraction property to increase emotion recognition;
• generate attribute feature maps using the facial identifier feature map, the attribute feature maps associated with facial attributes;
• calculate probability factors for the respective facial attributes of the attribute feature map, the probability factors identifying the respective facial attributes;
• classify an emotion based on the probability factors for the respective facial attributes;
• and
• generate a report indicative of the classified emotion.

An apparatus to perform multi-task recognition comprising:

• means for building layers of a neural network to generate a facial identifier feature map to generate a facial identifier feature map based on image data, the facial identifier feature map to identify a face within the image data, the means for building layers to build the layers by:
• reducing redundant filters corresponding to lower layers of the neural network;
• learning a residual function reference to a layer input;
• and
• broaden a network width and provide a multi-scale feature extraction property to increase emotion recognition;
• means for generating attribute feature maps based on the facial identifier feature map, the attribute feature maps associated with facial attributes;
• means for calculating probability factors for the respective facial attributes of the attribute feature map, the probability factors identifying the respective facial attributes;
• means for classifying an emotion based on the probability factors for the respective facial attributes;
• and
• means for generating a report to generate a report indicative of the classified emotion.