Unified Multi-Task Deep Learning Machine for Object Recognition

Solution Overview

Problem

Conventional approaches to face and object recognition require separate algorithms and non-neural network techniques for sequential tasks, leading to significant latency and resource consumption.

Innovation Solution

A unified architecture multi-task deep learning machine that integrates object detection, face alignment, and verification tasks using hierarchical convolutional neural networks, shared convolutional layers, and weighted loss generation to leverage shared computations and resources.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If separate algorithms are used for face detection and face verification tasks, then each task can be performed with dedicated optimization, but the system experiences significant latency and consumes excessive computing power and memory resources

Engineering Contradiction:

Improvetask performance accuracyVSAvoidoperational latency

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent merges face detection and face verification into a single unified neural network architecture. The network simultaneously performs both tasks by sharing common convolutional layers for feature extraction, then branching into separate detection and verification pathways. This integration eliminates the latency of sequential processing while maintaining dedicated optimization for each task through task-specific loss functions and output layers.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The unified neural network is designed with multi-functionality to handle both face detection and face verification within a single model. The shared backbone layers extract general facial features that serve both purposes, while task-specific modules handle detection bounding boxes and verification score generation respectively. This universal architecture reduces operational latency by processing both tasks in parallel rather than sequentially.

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Reliability

If separate algorithms are used for face detection and face verification tasks, then each task can be performed with dedicated optimization, but the system consumes excessive computing power and memory resources

Engineering Contradiction:

Improvetask performance accuracyVSAvoidcomputing power and memory consumption

Core Design Contradiction:

ReliabilityVSUse of energy by moving object

Solution Approach 1:

The patent merges face detection and face verification into a single unified neural network architecture. The network simultaneously performs both tasks by sharing common convolutional layers for feature extraction, then branching into separate detection and verification pathways. This integration eliminates the latency of sequential processing while maintaining dedicated optimization for each task through task-specific loss functions and output layers.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The unified neural network is designed with multi-functionality to handle both face detection and face verification within a single model. The shared backbone layers extract general facial features that serve both purposes, while task-specific modules handle detection bounding boxes and verification score generation respectively. This universal architecture reduces operational latency by processing both tasks in parallel rather than sequentially.

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Measurement precision

If non-neural network techniques are used for face cropping and alignment, then task-specific precision can be achieved, but significant computational resources and time are consumed

Engineering Contradiction:

Improveface alignment precisionVSAvoidcomputational efficiency

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The patent replaces traditional mechanical/non-neural network face cropping and alignment techniques with a neural network-based approach. The unified network learns optimal alignment transformations directly from data, substituting hand-crafted algorithms with a learned model that achieves comparable or superior precision while reducing computational overhead through shared feature extraction and integrated processing.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS11645869B2System and method for a unified architecture multi-task deep learning machine for object recognition

Publication Date: 2023.05.09 SAMSUNG ELECTRONICS CO LTD

AI summary

A system to recognize objects in an image includes an object detection network outputs a first hierarchical-calculated feature for a detected object. A face alignment regression network determines a regression loss for alignment parameters based on the first hierarchical-calculated feature. A detection box regression network determines a regression loss for detected boxes based on the first hierarchical-calculated feature. The object detection network further includes a weighted loss generator to generate a weighted loss for the first hierarchical-calculated feature, the regression loss for the alignment parameters and the regression loss of the detected boxes. A backpropagator backpropagates the generated weighted loss. A grouping network forms, based on the first hierarchical-calculated feature, the regression loss for the alignment parameters and the bounding box regression loss, at least one of a box grouping, an alignment parameter grouping, and a non-maximum suppression of the alignment parameters and the detected boxes.