Image Classification
A computer vision task that assigns one or more labels to an entire image, determining what category or categories the image belongs to.
Approaches
Modern image classification uses pre-trained vision models (ResNet, EfficientNet, ViT) fine-tuned on task-specific data. CLIP enables zero-shot classification by matching images against text descriptions.
Benchmarks
ImageNet: 1000-class classification with 1.2M images. CIFAR-10/100: Small-scale benchmarks. Top-1 accuracy on ImageNet exceeded human performance (95%+) by 2015 and continues to improve.