VectorMapNet: End-to-end Vectorized HD Map Learning

Authors: Yicheng Liu, Tianyuan Yuan, Yue Wang, Yilun Wang, Hang Zhao

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments show that Vector Map Net achieve strong map learning performance on both nu Scenes and Argoverse2 dataset, surpassing previous state-of-the-art methods by 14.2 m AP and 14.6m AP.
Researcher Affiliation Collaboration 1Shanghai Qi Zhi Institute 2Tsinghua University 3MIT 4Li Auto. Correspondence to: Hang Zhao <Zhao Hang0124@gmail.com>.
Pseudocode Yes Algorithm 1 The Algorithm of Discrete Fr echet Distance
Open Source Code No The paper does not contain any explicit statements about releasing source code for the methodology or links to a code repository.
Open Datasets Yes We conduct experiments on the nu Scenes (Caesar et al., 2020) and Argoverse2 (Wilson et al., 2021) dataset.
Dataset Splits Yes Argoverse2 We further conduct experiments on Argoverse2 (Wilson et al., 2021) dataset. Like nu Scenes, it contains 1000 logs (700, 150, 150 for training, validation and test set).
Hardware Specification Yes We train all our models on 8 GTX3090 GPUs for 110 epochs with a total batch size of 32.
Software Dependencies No The paper mentions software components like 'Res Net50', 'Point Net', 'Adam W optimizer', and 'Transformer', but does not provide specific version numbers for these or any other libraries or frameworks used (e.g., PyTorch, TensorFlow, CUDA versions).
Experiment Setup Yes We train all our models on 8 GTX3090 GPUs for 110 epochs with a total batch size of 32. We use Adam W (Loshchilov & Hutter, 2018) optimizer with a gradient clipping norm of 5.0. For the learning rate schedule, we use a step schedule that multiplies a learning rate by 0.1 at epoch 100 and has a linear warm-up period at the first 5000 steps. The dropout rate for all modules is 0.2, following the transformer s settings (Vaswani et al., 2017).