R4D: Utilizing Reference Objects for Long-Range Distance Estimation

Authors: Yingwei Li, Tiffany Chen, Maya Kabkab, Ruichi Yu, Longlong Jing, Yurong You, Hang Zhao

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on the two proposed datasets demonstrate the effectiveness and robustness of R4D by showing significant improvements compared to existing baselines.
Researcher Affiliation Industry Waymo LLC {ywli, yuhanc, kabkabm}@waymo.com
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper states that the Waymo Open Dataset Long-Range Labels will be made available publicly, but does not provide a concrete access statement or link for the source code of the R4D methodology itself.
Open Datasets Yes We re looking to make this dataset available publicly at waymo.com/open/download.
Dataset Splits Yes the Pseudo Long-Range KITTI Dataset contains 2,181 images with 4,233 vehicles, and 2,340 images with 4,033 vehicles in the training and validation sets, respectively.
Hardware Specification Yes We train R4D on an 8-core TPU with a total batch size of 32 images for 24 epochs.
Software Dependencies No The paper mentions software components like ResNet-50, ROIAlign, smooth L1, and Layer Normalization, but does not provide specific version numbers for any of them.
Experiment Setup Yes We train R4D on an 8-core TPU with a total batch size of 32 images for 24 epochs. The learning rate is initially set to 0.0005 and then decays by a factor of 10 at the 16th and 22nd epochs. We warm up (Goyal et al., 2017) the learning rate for 1,800 iterations.