This is code to fit per-pixel environment map with spherical Gaussian lobes, using LBFGS optimization. This code has been used in the following paper to generate ground-truth spherical Gaussian parameters.

• Li, Z., Shafiei, M., Ramamoorthi, R., Sunkavalli, K., & Chandraker, M. (2020). Inverse rendering for complex indoor scenes: Shape, spatially-varying lighting and svbrdf from a single image. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 2475-2484).

Please cite the paper if you find the code to be useful in your own project. Please refer to the papers for more implementation details.

To run the code, use the command python optimEnvSplit.py --cuda --dataRoot DATA, where DATA is the path to the synthetic dataset.