A Simple Baseline for Multi-Camera 3D Object Detection

Authors: Yunpeng Zhang, Wenzhao Zheng, Zheng Zhu, Guan Huang, Jiwen Lu, Jie Zhou

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

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive experiments on the 3D object detection benchmark of nu Scenes to demonstrate the effectiveness of Sim MOD and achieve competitive performance.
Researcher Affiliation Collaboration 1Phi Gent Robotics 2Department of Automation, Tsinghua University
Pseudocode No The paper does not contain any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes Code will be available at https://github.com/zhangyp15/Sim MOD.
Open Datasets Yes We conduct extensive experiments on the large-scale autonomous driving dataset nu Scenes (Caesar et al. 2020) to evaluate the proposed method.
Dataset Splits Yes These sequences are officially split into 700/150/150 ones for training, validation, and testing.
Hardware Specification Yes We train the model on 8 NVIDIA Ge Force RTX 3090 GPUs with per-GPU batch size as 1.
Software Dependencies No Sim MOD is implemented based on mmdetection3d. This only mentions one software component and does not provide version numbers for other key dependencies like Python, PyTorch, or CUDA.
Experiment Setup Yes The model is end-to-end trained with Adam W (Loshchilov and Hutter 2019) optimizer for 24 epochs, with the initial learning rate as 2e-4 and weight decay as 0.01. Multi-step learning rate decay is applied. The input resolution is 1600 900. ... We use the top-scored 600 proposals and filter low-scored proposals. The detection head contains six layers for iterative refinement.