Single-shot autofocus via adaptive foveation from global perception to local focus

• Proposes a single-shot autofocus network that jointly models global scene perception and local sharpness features. • Introduces an adaptive foveation mechanism to balance large-scale contextual information and fine-grained focus cues. • Employs sharpness-guided feature modulation to enhance informative regions during defocus regression. • Experiments on biological and industrial chip datasets demonstrate accurate and stable defocus estimation under varying defocus levels and texture complexity.