Computer Vision

Computer vision encompasses a wide range of tasks, each with different model architectures and training requirements.

• • Image classification: assigning a label to an entire image (e.g., "cat" or "invoice").
• • Object detection: locating and identifying multiple objects within an image with bounding boxes.
• • Semantic segmentation: labeling every pixel in an image with a class.
• • Optical character recognition (OCR): extracting text from images and documents.
• • Video understanding: tracking objects and events across frames over time.

For a decade, convolutional neural networks (CNNs) dominated computer vision by exploiting the spatial structure of images through learnable filters. Vision Transformers (ViT), introduced in 2020, adapted the self-attention mechanism from NLP to image patches, achieving state-of-the-art results and enabling tighter integration with language models.

Practical deployments of computer vision span industries: automated quality inspection in manufacturing, identity verification in fintech, receipt and invoice parsing in accounting, and visual search in e-commerce. Modern multimodal models like GPT-4V and Claude 3 allow businesses to query images with natural language without building custom vision pipelines.