English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Zendy Plus

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Zendy Tools

Zendy Open

Deep learning has recently achieved significant progress in trajectoryforecasting. However, the scarcity of trajectory data inhibits the data-hungrydeep-learning models from learning good representations. While maturerepresentation learning methods exist in computer vision and natural languageprocessing, these pre-training methods require large-scale data. It is hard toreplicate these approaches in trajectory forecasting due to the lack ofadequate trajectory data (e.g., 34K samples in the nuScenes dataset). To workaround the scarcity of trajectory data, we resort to another data modalityclosely related to trajectories-HD-maps, which is abundantly provided inexisting datasets. In this paper, we propose PreTraM, a self-supervisedpre-training scheme via connecting trajectories and maps for trajectoryforecasting. Specifically, PreTraM consists of two parts: 1) Trajectory-MapContrastive Learning, where we project trajectories and maps to a sharedembedding space with cross-modal contrastive learning, and 2) Map ContrastiveLearning, where we enhance map representation with contrastive learning onlarge quantities of HD-maps. On top of popular baselines such as AgentFormerand Trajectron++, PreTraM boosts their performance by 5.5% and 6.9% relativelyin FDE-10 on the challenging nuScenes dataset. We show that PreTraM improvesdata efficiency and scales well with model size.

PreTraM: Self-Supervised Pre-training via Connecting Trajectory and Map