Dynamic to Static Lidar Scan Reconstruction Using Adversarially Trained Auto Encoder

Authors: Prashant Kumar, Sabyasachi Sahoo, Vanshil Shah, Vineetha Kondameedi, Abhinav Jain, Akshaj Verma, Chiranjib Bhattacharyya, Vinay Vishwanath1836-1844

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our proposed approaches against baselines on three different datasets with the following goals: (1) In section 4, we evaluate our proposed approaches with adapted baseline models for the problem of DST for Li DAR, (2) In section 4, we evaluate our proposed approaches DSLR, DSLR-Seg, and DSLR-UDA, for Li DAR based SLAM. Experiments on simulated and real-world datasets show that DSLR gives at least a 4 improvement over adapted baselines.
Researcher Affiliation Collaboration 1Indian Institute of Science, Bangalore, India 2AMIDC Pvt Ltd, Bangalore, India 3Chennai Mathematical Institute, Chennai, India {prshnttkmr, ssahoo.iisc, vanshilshah, vineetha.knd92, abhinav98jain, akshajverma7}@gmail.com, chiru@iisc.ac.in, vinay@atimotors.com
Pseudocode No The paper does not contain any pseudocode or clearly labeled algorithm blocks.
Open Source Code Yes Code, Dataset and Appendix: https://dslrproject.github.io/dslr/
Open Datasets Yes We open-source 2 new datasets, CARLA-64, ARD16 (Ati Realworld Dataset) consisting of corresponding static-dynamic Li DAR scan pairs for simulated and real world scenes respectively. ... KITTI-64 dataset: To show results on a standard real-world dataset, we use the KITTI odometry dataset (Geiger, Lenz, and Urtasun 2012), which contains segmentation information and ground truth poses.
Dataset Splits No The paper mentions training and testing on datasets but does not provide explicit training/validation/test dataset splits with percentages or sample counts.
Hardware Specification No The paper does not explicitly describe the hardware (e.g., specific GPU or CPU models) used for running its experiments.
Software Dependencies No The paper mentions software components like DCGAN, U-Net, Open3D, and Cartographer, but does not provide specific version numbers for these tools or any other software dependencies.
Experiment Setup Yes We change the model architecture to a 40 512 grid instead of a 64 1024 grid. This is done to discard the outer circles in a Li DAR scan because it contains the most noise and have least information about the scene. The bottleneck dimension of our model is 160. ... 휆is a constant which is set to 0.01.