AdaPKC: PeakConv with Adaptive Peak Receptive Field for Radar Semantic Segmentation

Authors: Teng Li, Liwen Zhang, Youcheng Zhang, ZijunHu , Pengcheng Pi, Zongqing Lu, Qingmin Liao, Zhe Ma

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through experimental verification using various real-measured radar data (including publicly available low-cost millimeter-wave radar dataset for autonomous driving and self-collected Ku-band surveillance radar dataset), we found that the performance of Ada PKC-based models surpasses other So TA methods in RSS tasks.
Researcher Affiliation Collaboration 1Intelligent Science and Technology Academy of CASIC 2Shenzhen International Graduate School, Tsinghua University
Pseudocode Yes Algorithm 1: Voting-driven Multi-round Training
Open Source Code Yes The code is available at https://github.com/lihua199710/Ada PKC.
Open Datasets Yes CARRADA [20] dataset is recorded by a low-cost FMCW radar... CARRADA-RAC [32] dataset is derived from CARRADA... Ku RALS dataset is self-collected by a Kurz-under band (∼17GHz) surveillance Radar...
Dataset Splits Yes The dataset splits are the same as in [32, 19].
Hardware Specification Yes Frame rate is calculated on a workstation with an Intel(R) Xeon(R) Platinum 8255C CPU and a Tesla V100-SXM2 GPU. ...We train all these models on two NVIDIA-3090 GPUs...
Software Dependencies No The paper mentions using the Adam optimizer, but does not provide specific version numbers for key software components or libraries like PyTorch, TensorFlow, or Python.
Experiment Setup Yes The input sizes of RA, AD and RD views are 256 × 256, 256 × 64 and 256 × 64, respectively... The initial learning rate is 1e−4, and decays in a cosine manner by default. We train these models for 300 epochs with a batch size of 6.