VoGE: A Differentiable Volume Renderer using Gaussian Ellipsoids for Analysis-by-Synthesis

Authors: Angtian Wang, Peng Wang, Jian Sun, Adam Kortylewski, Alan Yuille

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Quantitative and qualitative experiment results show Vo GE outperforms So TA counterparts when applied to various vision tasks, e.g., object pose estimation, shape/texture fitting, and occlusion reasoning.
Researcher Affiliation Collaboration Angtian Wang1, Peng Wang2, Jian Sun2, Adam Kortylewski3, and Alan Yuille1 1 Johns Hopkins University 2Byte Dance Inc. 3Max Planck Institute for Informatics
Pseudocode No No pseudocode or algorithm blocks are provided in the paper.
Open Source Code Yes The code is available: https://github.com/Angtian/Vo GE.
Open Datasets Yes Dataset. Following Ne Mo, we evaluate pose estimation performance on the PASCAL3D+ dataset Xiang et al. (2014), the Occluded PASCAL3D+ dataset Wang et al. (2020b) and the Object Net3D dataset Xiang et al. (2016).
Dataset Splits No The PASCAL3D+ dataset contains objects in 12 man-made categories with 11045 training images and 10812 testing images.
Hardware Specification No Using CUDA from NVIDIA et al. (2022), we implement Vo GE with both forward and backward function. The CUDA-Vo GE is packed as an easy-to-use autogradable Py Torch API.
Software Dependencies Yes Using CUDA from NVIDIA et al. (2022), we implement Vo GE with both forward and backward function. The CUDA-Vo GE is packed as an easy-to-use autogradable Py Torch API. NVIDIA, P eter Vingelmann, and Frank H.P. Fitzek. Cuda, release: 11.2.89, 2022. URL https: //developer.nvidia.com/cuda-toolkit.
Experiment Setup Yes Experiment Details. Following the experiment setup in Ne Mo, we train the feature extractor 800 epochs with a progressive learning rate. During inference, for each image, we sample 144 starting poses and optimizer 300 steps via an ADAM optimizer.