RefRef: A Dataset and Benchmark for Reconstructing Refractive and Reflective Objects
The episode discusses 'RefRef: A Dataset and Benchmark for Reconstructing Refractive and Reflective Objects,' a paper introducing a synthetic dataset and benchmark to test AI's ability to reconstruct scenes with refractive and reflective objects. The hosts explore the authors' methods, including an oracle method for perfect light path calculation, and R3F, a relaxation method that estimates geometry without needing perfect ground truth.