The paper introduces Pixel-Aligned Implicit Function (PIFu), a novel representation for 3D deep learning that aligns pixel-level information from 2D images with the global context of 3D objects. PIFu is designed to digitize highly detailed clothed humans from a single or multiple input images, inferring both 3D surface and texture. The method can handle intricate shapes, such as hairstyles and clothing, and captures fine-scale details in unseen regions. PIFu is memory-efficient, can handle arbitrary topology, and is spatially aligned with the input image. The approach is extended to multi-view input images, improving reconstruction accuracy. Experiments on real-world datasets demonstrate state-of-the-art performance in single-view and multi-view reconstruction, outperforming previous methods. The project website is available at <https://shunsukesaito.github.io/PIFu/>.The paper introduces Pixel-Aligned Implicit Function (PIFu), a novel representation for 3D deep learning that aligns pixel-level information from 2D images with the global context of 3D objects. PIFu is designed to digitize highly detailed clothed humans from a single or multiple input images, inferring both 3D surface and texture. The method can handle intricate shapes, such as hairstyles and clothing, and captures fine-scale details in unseen regions. PIFu is memory-efficient, can handle arbitrary topology, and is spatially aligned with the input image. The approach is extended to multi-view input images, improving reconstruction accuracy. Experiments on real-world datasets demonstrate state-of-the-art performance in single-view and multi-view reconstruction, outperforming previous methods. The project website is available at <https://shunsukesaito.github.io/PIFu/>.