A Python toolkit for working with the nuScenes and nuImages autonomous driving datasets, providing data loading, visualization, and evaluation utilities.
nuScenes devkit is a Python software development kit for the nuScenes and nuImages autonomous driving datasets. It provides data loaders, visualization tools, and evaluation scripts to help researchers and engineers work with these large-scale, multimodal datasets for developing and benchmarking perception, prediction, and tracking algorithms for self-driving vehicles.
Autonomous driving researchers, computer vision engineers, and machine learning practitioners who need to train, evaluate, or experiment with models on the nuScenes or nuImages datasets.
It offers the official, standardized interface and evaluation code for the nuScenes ecosystem, ensuring reproducibility and fair comparison in benchmarks. The toolkit abstracts away the complexity of handling multimodal sensor data and annotations, accelerating research and development.
The devkit of the nuScenes dataset.
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Provides unified Python classes (NuScenes, NuImages) to load multimodal sensor data, annotations, and expansions like Panoptic nuScenes, lidarseg, CAN bus, and maps, as shown in the setup instructions for each dataset.
Includes official benchmarking code for 3D detection, tracking, segmentation, and prediction tasks, ensuring reproducibility and fair comparisons in challenges, referenced in the evaluation readme and tutorial links.
Offers Jupyter notebook tutorials for data visualization and exploration, plus detailed schema and annotator instructions, making it easier for users to get started quickly.
Seamlessly supports dataset expansions such as map layers and panoptic labels through a consistent API, with dedicated setup steps and tutorials for each expansion.
Requires downloading multiple large archives, adhering to a strict folder structure, and managing dependencies, as detailed in the setup sections for nuScenes and nuImages, which can be time-consuming.
Only tested for Python 3.9 and 3.12, with strict pip requirements specified in the changelog and installation docs, limiting compatibility with other environments.
Admits to persistent problems like incorrect traffic light poses and ego pose inaccuracies in certain maps, as listed in the Known Issues section, which can affect model training or evaluation.