Patch Based Inference

Patch-based inference is a technique that processes images or other data by dividing them into smaller patches, analyzing each individually, and then aggregating the results. Current research focuses on optimizing this approach for efficiency and accuracy, particularly on resource-constrained devices like microcontrollers, through methods like mixed-precision quantization and value-driven patch classification. This approach improves the efficiency of deep learning models, especially in applications like digital pathology and pavement distress detection, where high-resolution images are common, enabling faster processing and wider accessibility of these powerful tools.