MERL BRDF Database (cdfg.csail.mit.edu)

The MERL BRDF Database is a curated collection of densely sampled bidirectional reflectance distribution functions (BRDFs) for 100 real-world materials. Each entry encodes a measured 4D reflectance function (incoming/outgoing angles) and the distribution is distributed with example reader code; note that the parameterization of the half-angle (theta-half) in the files has changed since earlier releases. The dataset is intended for research and academic use only and should be cited as Matusik et al., “A Data‑Driven Reflectance Model,” ACM TOG (2003) when used. For the AI/ML community, MERL is a high-value ground truth resource for training and evaluating models that handle material appearance: neural and differentiable rendering, inverse-rendering and reflectance estimation, relighting, and data-driven physically based rendering. Because the BRDFs are densely measured rather than analytic fits, they provide realistic, high-fidelity targets for supervised learning, benchmarking, and hybrid model-based/data-driven pipelines. Practical notes: use the provided sample code to parse the files and account for the theta-half parameterization change, and respect the dataset’s research-only license and citation requirement when publishing results.