SIGGRAPH 2025 Significant New Researcher Award: Mildenhall
Awardee(s):
Award:
-
Significant New Researcher Award
Description:
ACM SIGGRAPH is pleased to present the 2024 Significant New Researcher Award to Ben Mildenhall and Pratul Srinivasan for their outstanding contributions to new representations for 3D graphics, neural rendering, novel view synthesis, and generative models of 3D scenes.
Ben and Pratul’s research centers on new representations and algorithms for capturing, rendering, and generating 3D scenes. Before their work, computer graphics had largely settled on explicit surface-based representations, such as triangles and subdivision surfaces together with textures and BRDF for material appearance. However, these representations made it challenging to capture fully realistic models of real scenes and to synthesize scenes. They were also poorly suited to new deep learning methodology because their hardness and arbitrary topology did not interact well with optimization techniques and derivative computations.
Ben and Pratul introduced radically new representations and algorithms for the neural age that have dramatically improved our ability to capture the real world and synthesize 3D scenes. Their work on the Neural Radiance Field (NeRF) introduced a fundamentally new way of representing 3D scenes, combining volume rendering and neural networks as the scene representation itself. They also analyzed and demonstrated the importance of Fourier features to achieve high-resolution representations. They then introduced multiple new representations to scale up neural representations to richer scenes and higher-fidelity reconstructions, including Mip-NeRF, Mip-NeRF360, Block-NeRF, and Zip-NeRF.
They also made important contributions to computational imaging using advanced priors, in the areas of black hole imaging for Pratul and lensless diffuser cameras for Ben.
Ben and Pratul also spearheaded the area of generative AI for 3D. In DreamFusion, Ben and co-authors were the first to demonstrate the use of 2D diffusion models to train the synthesis of 3D models from text prompts, sidestepping the need for 3D training data. In CAT3D, Pratul and co-authors introduced a multi-view latent diffusion model to generate novel views of a scene.
In summary, Ben Mildenhall and Pratul Srinivasan revolutionized the full gamut of 3D graphics from the capture of real scenes with neural representations, to computational imaging, all the way to generative models for the synthesis of 3D scenes.
Ben Mildenhall received a B.S from Stanford University in 2015 and a PhD from UC Berkeley in 2020 advised by Ren Ng. During his PhD, he interned with Jon Barron at Google Research and with Rodrigo Ortiz-Cayon and Abhishek Kar at Fyusion. He was a research scientist at Google Research before cofounding World Labs.
Pratul Srinivasan received a B.S.E. from Duke University in 2014 and a PhD from UC Berkeley in 2020, where he was supervised by Ren Ng and Ravi Ramamoorthi. During his PhD, he interned at Google Research with Jon Barron and Noah Snavely. He is currently a research scientist at Google DeepMind.
