Computer VisionAcademicOpenSourceVERIFIED

CNN (Convolutional Neural Network)

by Academic (multiple contributors) • Founded 1980

A Convolutional Neural Network (CNN or ConvNet) is a class of deep neural networks designed to automatically and adaptively learn spatial hierarchies of features from input images and other grid-like data. By using convolutional layers, pooling, and non-linear activations, CNNs excel at recognizing patterns such as edges, textures, and objects with far fewer parameters than fully connected networks. They are foundational to modern computer vision and have enabled breakthroughs in image classification, detection, segmentation, and many other perception tasks.

Key Features

•Convolutional layers that learn local receptive fields and spatially shared filters
•Hierarchical feature extraction from low-level edges to high-level object parts
•Parameter sharing and sparse connectivity, reducing model size and overfitting risk
•Pooling/subsampling layers for translation invariance and dimensionality reduction
•Support for multi-channel inputs (e.g., RGB images, feature maps)
•Compatibility with GPUs and accelerators for efficient large-scale training
•Extensibility to 1D and 3D convolutions for sequences, audio, and volumetric data

Use Cases

•Image classification (e.g., object recognition in photos)
•Object detection and localization (e.g., autonomous driving, surveillance)
•Semantic and instance segmentation (e.g., medical imaging, scene understanding)
•Face recognition and verification
•Optical character recognition (OCR) and document analysis
•Medical image analysis (e.g., tumor detection in MRI/CT scans)
•Image super-resolution, denoising, and enhancement
•Content moderation and visual search in social media and e-commerce