PhysAvatar is a novel framework that combines inverse rendering with inverse physics to automatically estimate the shape and appearance of a human from multi-view video data along with the physical parameters of the fabric of their clothes. The framework uses a mesh-aligned 4D Gaussian technique for spatio-temporal mesh tracking and a physically based inverse renderer to estimate intrinsic material properties. PhysAvatar integrates a physics simulator to estimate the physical parameters of the garments using gradient-based optimization. This enables the creation of high-quality novel-view renderings of avatars dressed in loose-fitting clothes under motions and lighting conditions not seen in the training data. The key contributions of PhysAvatar include the introduction of a new inverse rendering paradigm for avatars that incorporates the physics of loose garments, a pipeline that includes accurate and efficient mesh reconstruction and tracking using 4D Gaussians, automatic optimization of the garments' physical material properties, and accurate appearance estimation using physically based inverse rendering. PhysAvatar demonstrates a novel path to learning physical scene properties from visual observations by incorporating inverse rendering with physics constraints. The method is evaluated on the ActorHQ dataset and shows superior performance in terms of geometry accuracy and appearance modeling compared to existing methods. PhysAvatar can facilitate a wide range of applications, such as animation, relighting, and redressing. The method is also capable of being exported into traditional computer graphics pipelines, enabling post-processing artistic modifications. The framework is implemented using a combination of inverse rendering and physics-based simulation, and is evaluated on various metrics including PSNR, SSIM, and LPIPS. The results show that PhysAvatar outperforms existing methods in terms of geometry accuracy and appearance modeling. The method is also capable of handling challenging motions from Mixamo and can be integrated with computer graphics software tools. The framework is limited by the need for manual garment segmentation and mesh UV unwrapping, and the absence of a perfect underlying human body for garment simulation. Future work includes the development of more advanced parametric models and the adaptation of the method to sparse-view settings.PhysAvatar is a novel framework that combines inverse rendering with inverse physics to automatically estimate the shape and appearance of a human from multi-view video data along with the physical parameters of the fabric of their clothes. The framework uses a mesh-aligned 4D Gaussian technique for spatio-temporal mesh tracking and a physically based inverse renderer to estimate intrinsic material properties. PhysAvatar integrates a physics simulator to estimate the physical parameters of the garments using gradient-based optimization. This enables the creation of high-quality novel-view renderings of avatars dressed in loose-fitting clothes under motions and lighting conditions not seen in the training data. The key contributions of PhysAvatar include the introduction of a new inverse rendering paradigm for avatars that incorporates the physics of loose garments, a pipeline that includes accurate and efficient mesh reconstruction and tracking using 4D Gaussians, automatic optimization of the garments' physical material properties, and accurate appearance estimation using physically based inverse rendering. PhysAvatar demonstrates a novel path to learning physical scene properties from visual observations by incorporating inverse rendering with physics constraints. The method is evaluated on the ActorHQ dataset and shows superior performance in terms of geometry accuracy and appearance modeling compared to existing methods. PhysAvatar can facilitate a wide range of applications, such as animation, relighting, and redressing. The method is also capable of being exported into traditional computer graphics pipelines, enabling post-processing artistic modifications. The framework is implemented using a combination of inverse rendering and physics-based simulation, and is evaluated on various metrics including PSNR, SSIM, and LPIPS. The results show that PhysAvatar outperforms existing methods in terms of geometry accuracy and appearance modeling. The method is also capable of handling challenging motions from Mixamo and can be integrated with computer graphics software tools. The framework is limited by the need for manual garment segmentation and mesh UV unwrapping, and the absence of a perfect underlying human body for garment simulation. Future work includes the development of more advanced parametric models and the adaptation of the method to sparse-view settings.