The nuScenes dataset is a large-scale autonomous driving dataset. The dataset has 3D bounding boxes for 1000 scenes collected in Boston and Singapore. Each scene is 20 seconds long and annotated at 2Hz. This results in a total of 28130 samples for training, 6019 samples for validation and 6008 samples for testing. The dataset has the full autonomous vehicle data suite: 32-beam LiDAR, 6 cameras and radars with complete 360° coverage. The 3D object detection challenge evaluates the performance on 10 classes: cars, trucks, buses, trailers, construction vehicles, pedestrians, motorcycles, bicycles, traffic cones and barriers.
1,615 PAPERS • 20 BENCHMARKS
IndustReal is an ego-centric, multi-modal dataset where 27 participants are challenged to perform assembly and maintenance procedures on a construction-toy car. The dataset is annotated for action recognition, assembly state detection, and procedure step recognition. IndustReal includes 38 execution errors in a total of 84 videos, with 14 exclusive to validation and test sets and therefore suitable for testing the robustness of algorithms against unseen errors in procedural tasks. IndustReal offers open-source 3D models for all parts to promote the use of synthetic data for scalable approaches on this dataset, as well as reproducibility. All assembly parts used in the dataset are 3D printed. This ensures reproducibility and future availability of the model and allows for growth via community effort.
2 PAPERS • 2 BENCHMARKS
Understanding comprehensive assembly knowledge from videos is critical for futuristic ultra-intelligent industry. To enable technological breakthrough, we present HA-ViD – an assembly video dataset that features representative industrial assembly scenarios, natural procedural knowledge acquisition process, and consistent human-robot shared annotations. Specifically, HA-ViD captures diverse collaboration patterns of real-world assembly, natural human behaviors and learning progression during assembly, and granulate action annotations to subject, action verb, manipulated object, target object, and tool. We provide 3222 multi-view and multi-modality videos, 1.5M frames, 96K temporal labels and 2M spatial labels. We benchmark four foundational video understanding tasks: action recognition, action segmentation, object detection and multi-object tracking. Importantly, we analyze their performance and the further reasoning steps for comprehending knowledge in assembly progress, process effici
1 PAPER • NO BENCHMARKS YET