Revisiting 3D Object Detection From an Egocentric Perspective

Authors: Boyang Deng, Charles R Qi, Mahyar Najibi, Thomas Funkhouser, Yin Zhou, Dragomir Anguelov

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments on the large-scale Waymo Open Dataset show that SDE better reflects the impact of detection quality on the ego-agent s safety compared to Io U; and the estimated contours from Star Poly consistently improve the egocentric detection quality over recent 3D object detectors.
Researcher Affiliation Industry Boyang Deng Charles R. Qi Mahyar Najibi Thomas Funkhouser Yin Zhou Dragomir Anguelov Waymo LLC Google Research
Pseudocode No The paper does not contain any clearly labeled 'Pseudocode' or 'Algorithm' blocks, nor does it present structured steps in a code-like format.
Open Source Code No The code and the data are proprietary. We plan to release the code of computing the metrics upon acceptance.
Open Datasets Yes All analysis is based on the Waymo Open Dataset [53] validation set.
Dataset Splits Yes All analysis is based on the Waymo Open Dataset [53] validation set.
Hardware Specification Yes We include this information in the supplementary material for our proposed Star Poly model.
Software Dependencies No The paper mentions using a 'Point Net [40] model' and 'Adam' optimizer but does not specify version numbers for any software dependencies or libraries required for reproduction.
Experiment Setup Yes We use a resolution n = 256 for all following experiments. ( d1, ..., dn) is uniformly sampled from the boundary of a square. During training, γ and β are both set to 0.1, which is determined by a grid search over the hyperparameters. Please refer to the supplementary material for more model details.