Language Prompt for Autonomous Driving

Authors: Dongming Wu, Wencheng Han, Yingfei Liu, Tiancai Wang, Cheng-Zhong Xu, Xiangyu Zhang, Jianbing Shen

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we conduct a set of experiments on our proposed benchmark Nu Prompt. For implementation details, we follow the settings of PF-Track (Pang et al. 2023). ... Experiments show that our Prompt Track achieves impressive performance on Nu Prompt.
Researcher Affiliation Collaboration 1 Beijing Institute of Technology, 2 SKL-IOTSC, CIS, University of Macau, 3 MEGVII Technology, 4 Beijing Academy of Artificial Intelligence EMAIL, EMAIL, EMAIL
Pseudocode No The paper describes the methodology using text and mathematical equations, but it does not include a clearly labeled pseudocode block or algorithm.
Open Source Code No The paper does not contain an explicit statement about releasing code or a link to a code repository for the described methodology.
Open Datasets Yes Our Nu Prompt is built on one of the most popular datasets for multi-view 3D object detection, nu Scenes (Caesar et al. 2019). ... To advance the research of prompt learning in driving scenarios, we propose a new large-scale benchmark, named Nu Prompt.
Dataset Splits Yes The Nu Prompt contains a total of 850 videos along with language prompts. Following nu Scenes (Caesar et al. 2019), we split Nu Prompt into training and validation set, which contain 700 videos and 150 videos, respectively.
Hardware Specification Yes For inference speed, we test Prompt Track using VOV backbone on one Nividia A100 GPU over the validation set. Prompt Track achieves 7.7 FPS.
Software Dependencies No The paper mentions GPT3.5 (Open AI 2023) for prompt generation, but it does not specify software dependencies with version numbers for the experimental setup of Prompt Track.
Experiment Setup No For implementation details, we follow the settings of PF-Track (Pang et al. 2023). ... Following the work (Pang et al. 2023), the number of fixed queries is set to 500. ... The settings of their weights λs follow PF-Track (Pang et al. 2023).