Achieving physically consistent image editing remains a significant challenge in computer vision. Existing image editing methods typically rely on neural networks, which struggle to accurately handle shadows and refractions. Conversely, physics-based inverse rendering often requires multi-view optimization, limiting its practicality in single-image scenarios. In this paper, we propose Materialist , a neural-initialized physically based rendering pipeline for single-image inverse rendering. Unlike previous hybrid methods that use physics to guide neural generation, our method leverages neural networks to predict initial material properties, which are then rigorously optimized via progressive differentiable rendering. Our approach enables a range of applications, including material editing, object insertion, and relighting, while also introducing an effective method for editing material transparency via ray-traced refraction without requiring full scene geometry. Furthermore, our envmap estimation method also achieves competitive performance, further enhancing the accuracy of image editing task. Experiments demonstrate strong performance across synthetic and real-world datasets, excelling even on challenging out-of-domain images.
Quantifying anisotropy in a turbid medium is critical for applications ranging from biomedical diagnostics to food quality assessment. Using a polarization camera and irradiation by a linearly polarized beam, we present a noninvasive method for estimating material anisotropy by analyzing depolarization ratios. To validate our approach, we perform a 3D microstructural analysis of mozzarella cheese samples to obtain ground-truth anisotropy levels using a metric based on structure tensor scale-space analysis of computed tomography (CT) scans. Our experiments with mozzarella samples of different anisotropy levels demonstrate a stronger correlation with CT-derived anisotropy and greater robustness as compared with oblique incidence reflectometry, which is limited by surface artifacts and sensitivity to sample orientation. Our method maintains high accuracy with simpler instrumentation and without sample rotation even when samples exhibit surface variations (non-flatness or texture). This establishes polarization imaging as a robust platform for noninvasive anisotropy characterization with potential applications in industrial quality control and biomedical diagnostics.