publications

2023

PAPR: Proximity Attention Point Rendering
Yanshu Zhang*, Shichong Peng*, Alireza Moazeni, Ke Li
Neural Information Processing Systems (NeurIPS) - Spotlight
[Paper] [Project Page] [Code]

Learning accurate and parsimonious point cloud representations of scene surfaces from scratch remains a challenge in 3D representation learning. We propose Proximity Attention Point Rendering (PAPR), a novel method that consists of a point-based scene representation and a differentiable renderer. PAPR effectively learns point cloud positions to represent the correct scene geometry and captures fine texture details while using only a parsimonious set of points.

2022

CHIMLE: Conditional Hierarchical IMLE for Multimodal Conditional Image Synthesis
Shichong Peng, Alireza Moazeni, Ke Li
Neural Information Processing Systems (NeurIPS) [Paper] [Project Page] [Code]

A new method to generate diverse output images from the same input image. We leverage Implicit Maximum Likelihood Estimation (IMLE) which can overcome mode collapse fundamentally. CHIMLE significantly outperforms the prior best IMLE, GAN and diffusion-based methods in terms of image fidelity and mode coverage across multiple image-to-image translation tasks.