LION: Latent Point Diffusion Models for 3D Shape Generation

Authors: xiaohui zeng, Arash Vahdat, Francis Williams, Zan Gojcic, Or Litany, Sanja Fidler, Karsten Kreis

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimentally, LION achieves state-of-the-art generation performance on multiple Shape Net benchmarks.
Researcher Affiliation Collaboration 1NVIDIA 2University of Toronto 3Vector Institute
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No We will release code and instructions to reproduce all experiments upon acceptance of the manuscript. The internal guidelines of our institution prevent us from releasing code at this stage.
Open Datasets Yes To compare LION against existing methods, we use Shape Net [104], the most widely used dataset to benchmark 3D shape generative models.
Dataset Splits Yes Following previous works [31, 46], we train on three categories: airplane, chair, car. Also like previous methods, we primarily rely on Point Flow s [31] dataset splits and preprocssing.
Hardware Specification Yes All experiments are performed on NVIDIA DGX servers with NVIDIA A100 GPUs.
Software Dependencies No The paper mentions several software tools and libraries used (e.g., PyTorch, Mit Suba renderer), but does not provide specific version numbers for these dependencies.
Experiment Setup Yes Our LION models use a batch size of 256 for all experiments. The encoder and decoder were trained with a learning rate of 1e-4 for 100 epochs, while the latent DDMs were trained with a learning rate of 2e-4 for 500 epochs. We use Adam optimizer...