And Lifan! Apologies that part of the text somehow dropped when including @ in Bluesky. 🥲🥲

Thank you! 😄

👺 By the way, we spent a lot of effort curating and will release a large-scale indoor dataset (∼14k scenes) with diverse floorplans, geometries, and textures, serving as a benchmark for learning light transport.

👾 We demonstrate the versatility of our framework by reusing a shared pre-trained encoder while introducing task-specific decoders.

👏 This is our team work: with my mentor Marco @marcosalvi.bsky.social , labmates Mukund, Cheng, NV folks Bart @bartwr.bsky.social , and Tzu-Mao @tzumaoli.bsky.social & Ravi!

It takes as input the 3D scene assets—geometry, materials, and lighting represented as a point cloud—and encodes the light interaction into a 3D latent space, with codes anchored at the sampled scene points. The model is independent of both viewpoint and resolution!

Building on the analogy of light transport operator and Attention, our fully scalable model learns a generalizable light transport 3D embedding that captures global illumination.