TopoLogic: An Interpretable Pipeline for Lane Topology Reasoning on Driving Scenes
Authors: Yanping Fu, Wenbin Liao, Xinyuan Liu, Hang Xu, Yike Ma, Yucheng Zhang, Feng Dai
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on the mainstream benchmark Open Lane-V2 for topology reasoning task indicate that our method significantly outperforms existing state-of-the-art methods, especially in lane topology metric. |
| Researcher Affiliation | Academia | Yanping Fu Institute of Computing Technology, Chinese Academy of Sciences University of Chinese Academy of Sciences fuyanping23s@ict.ac.cn Wenbin Liao Institute of Computing Technology, Chinese Academy of Sciences University of Chinese Academy of Sciences liaowenbin23z@ict.ac.cn Xinyuan Liu Institute of Computing Technology, Chinese Academy of Sciences University of Chinese Academy of Sciences liuxinyuan21s@ict.ac.cn Hang Xu Hangzhou Dianzi University hxu@hdu.edu.cn Yike Ma Institute of Computing Technology, Chinese Academy of Sciences ykma@ict.ac.cn Yucheng Zhang Institute of Computing Technology, Chinese Academy of Sciences zhangyucheng@ict.ac.cn Feng Dai Institute of Computing Technology, Chinese Academy of Sciences fdai@ict.ac.cn |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code is released at https://github.com/Franpin/Topo Logic. |
| Open Datasets | Yes | We have evaluated Topo Logic on the Open Lane-V2 [34] , which is currently the only large-scale perception and topology reasoning dataset devised for autonomous driving scenarios. |
| Dataset Splits | Yes | We have evaluated Topo Logic on the Open Lane-V2 [34] , which is currently the only large-scale perception and topology reasoning dataset devised for autonomous driving scenarios. This dataset was developed by Argogorse2 [35] and nu Scenes [36] respectively. It provides annotations for both lane centerline tasks and lane segment detection tasks. Open Lane-V2 consists of two subsets: subset_A and subset_B, each comprising 1000 scenes with 2Hz multi-view images and annotations. |
| Hardware Specification | Yes | All experiment is trained for 24 epochs on 8 NVIDIA RTX 3090 GPUs with a batch size of 16. |
| Software Dependencies | No | The paper mentions software components like 'Res Net-50', 'FPN', 'Deformable DETR', and 'Adam W optimizer', but does not provide specific version numbers for these or other ancillary software. |
| Experiment Setup | Yes | All images are resized into the same resolution of 1550 2048. ... The number of query is set to 200. ... For lane detection loss Ldetl, the weight of the classification part is 1.5, and the weight of the regression part is 0.025. ... α is initialized to 2, λ is initialized to 0.2. ... We train Topo Logic utilizing the Adam W optimizer [43] with a weight decay of 0.01 with an initial learning rate of 2 10 4 and employ a cosine annealing schedule for the learning rate. All experiment is trained for 24 epochs on 8 NVIDIA RTX 3090 GPUs with a batch size of 16. |