Exploiting Continuous Motion Clues for Vision-Based Occupancy Prediction

Authors: Haoran Xu, Peixi Peng, Xinyi Zhang, Guang Tan, Yaokun Li, Shuaixian Wang, Luntong Li

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on several benchmarks demonstrate that CMOP outperforms stateof-the-art baselines.
Researcher Affiliation Collaboration Haoran Xu1,2, Peixi Peng2,3*, Xinyi Zhang1,4, Guang Tan1*, Yaokun Li1, Shuaixian Wang1, Luntong Li2 1School of Intelligent Systems Engineering, Shenzhen Campus of Sun Yat-sen University 2Peng Cheng Laboratory 3School of Electronic and Computer Engineering, Shenzhen Graduate School, Peking University 4Nio Inc.
Pseudocode No The paper describes the methodology using prose and mathematical equations but does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes More detailed settings can be found in the supplementary material and the publicly available code repository1. 1https://github.com/kyoran/CMOP
Open Datasets Yes The nu Scenes benchmark (Caesar et al. 2020) along with dense occupancy ground truth (GT) labels (Wei et al. 2023b) are used for evaluation.
Dataset Splits Yes Following the official dataset split rule (Wei et al. 2023b), 700 and 150 scenes are employed for the training and testing phases, respectively.
Hardware Specification Yes All experiments are conducted on 4 Nvidia A100s, and deployed on the Nvidia Jetson AGX Orin.
Software Dependencies No The paper mentions using OpenCV for optical flow, but it does not provide specific version numbers for the software dependencies used in their implementation, such as Python, PyTorch, or CUDA versions.
Experiment Setup Yes The downsampled factor F of optical flow is set to 3 and the dimensions for the latent motion features Nmot is set to 100. The abstracted occupancy feature map is fixed at 32 9 9 1. The tradeoff coefficients of losses λsep,fuse are initially established at 0.4, and 0.2, respectively. To ensure efficient optimization, λsep is reduced progressively with a step-wise decay rate of 0.99998. Conversely, λfuse increases, maintaining the sum of all coefficients at 1.