Distortion and Uncertainty Aware Loss for Panoramic Depth Completion

Authors: Zhiqiang Yan, Xiang Li, Kun Wang, Shuo Chen, Jun Li, Jian Yang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments show the superiority of our method over standard loss functions, reaching the state of the art.
Researcher Affiliation Academia 1PCALab, School of Computer Science and Engineering, Nanjing University of Science and Technology, China 2RIKEN Center for Advanced Intelligence Project, Japan. Correspondence to: Shuo Chen <shuo.chen.ya@riken.jp>, Jun Li <junli@njust.edu.cn>.
Pseudocode No No pseudocode or algorithm blocks were found in the paper.
Open Source Code No The paper does not explicitly state that source code is released or provide a link to a code repository.
Open Datasets Yes Following M3PT (Yan et al., 2022), we train the model on Matterport3D (Albanis et al., 2021) and 3D60 (Zioulis et al., 2019) datasets with 512 256 resolution.
Dataset Splits No Matterport3D is composed of 7,907 RGB-D panoramas, 5,636 for training and 1,527 for testing. For 3D60, there are 6,669 RGB-D pairs for training and 1,831 for testing.
Hardware Specification Yes The whole training process is implemented on Pytorch with a single NVIDIA TITAN V GPU.
Software Dependencies No The paper mentions 'Pytorch' but does not specify its version or other software dependencies with version numbers.
Experiment Setup Yes Adam W optimizer is used with β1 = 0.9, β2 = 0.999 and weight decay 0.05. We train the model for 80 epoches with batch size 16 and initial learning rate 5 10 4, which drops by half every 20 epoches. Color jittering and random horizontal flip are used. µ and η are 80 and 0.5 respectively.