Microscopy has long been essential to biomedical research, enabling detailed analyses of complex samples. Fourier ptychographic microscopy (FPM), a computational imaging technique, provides high-resolution, wide-field images without requiring extensive hardware modifications. However, current FPM algorithms struggle with samples exhibiting depth variations, such as tilted or 3-dimensional (3D) objects. The limited depth of field (DoF) leads to images with only focal-plane areas in sharp focus, while regions outside appear blurred. To address this limitation, we propose an all-in-focus FPM algorithm using physics-informed 3D neural representations to reconstruct sharp, wide-field images of 3D objects under limited DoF. Unlike previous methods, our approach samples the full depth range to create a 3D feature volume that incorporates spatial and depth information.







