Face Detection Model Using Multi-Stage CNN Feature Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional face detection methods struggle to accurately recognize small-scale faces due to insufficient feature resolution and semantic information, as the feature extraction process in these methods typically results in lower resolution features by the final stage of convolutional neural networks, which cannot meet the requirements for detecting small-scale faces.
Innovation Solution
The proposed method uses a face detection model with multiple stages of convolutional neural networks, where features from adjacent stages are fused to enhance both resolution and semantic information, allowing for accurate detection of small-scale faces by utilizing projection features from both a first convolutional neural network with high resolution and a second network with more semantic information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional face detection methods use multi-stage convolutional neural networks for feature extraction, then the processing speed and general detection capability are improved, but the feature resolution deteriorates in later stages making small-scale face detection inaccurate
Solution Approach 1:
The patent segments the feature extraction process into multiple stages with different resolutions. It uses feature maps from different stages (P3, P4, P5) and applies selective processing: high-resolution features from earlier stages are preserved for small-scale faces, while lower-resolution semantic features from later stages are used for larger faces. This segmentation allows the system to maintain appropriate resolution for each detection task.
Solution Approach 2:
The patent introduces a new dimension by creating an additional processing path that preserves high-resolution features from earlier network stages. Instead of simply downsampling through all stages, it maintains parallel feature representations at different resolution levels and selectively combines them, adding a resolution dimension to the feature space.
2Loss of information
If the face detection model uses multiple stages of convolutional neural networks, then the semantic information extraction is improved, but the feature resolution for small-scale faces deteriorates
Solution Approach 1:
The patent merges features from multiple network stages by concatenating feature maps from different levels (P3, P4, P5). This combining approach preserves semantic information from deeper stages while maintaining resolution information from shallower stages, creating a comprehensive feature representation that contains both types of information simultaneously.
Solution Approach 2:
The patent applies local quality by treating different spatial regions and scale ranges differently. For small-scale faces, it prioritizes high-resolution features from earlier stages. For larger faces, it utilizes semantic-rich features from later stages. This localized adaptation of feature quality ensures optimal performance for each scale category.
3Use of energy by moving object
If conventional methods downsample features through all convolutional stages, then computational efficiency is improved, but detection precision for small-scale faces deteriorates
Solution Approach 1:
The patent introduces dynamic feature selection based on the detected face scale. The system adaptively chooses which feature stages to utilize: for small-scale faces, it dynamically selects and combines features from earlier high-resolution stages, while for large-scale faces, it uses features from later semantic stages. This dynamic adaptation optimizes computational resources for each specific detection case.
Data Source
AI summary
Embodiments of this disclosure disclose a face detection method, apparatus, and device, and a storage medium. A face candidate region in a to-be-detected image is determined according to a face detection model including a plurality of stages of convolutional neural networks (CNNs). A small-scale face is determined according to a size parameter, and a particular stage CNN corresponding to the size parameter is selected to perform the face detection. A CNN in different stage is adapted to a different size parameter or a size parameter range.