argoverse2

A framework for unified end-to-end autonomous driving datasets processing

AV2 Scenario Mining Challenge on CVPR Workshop 2026

Interactive 3D visualization and debugging tool for multi-object tracking across autonomous driving datasets — using a universal scene format for LiDAR, cameras, GT/tracker comparison, and MOT diagnostics.

Transformer map-matching localization in PyTorch: match detected landmarks against an HD map, solve the SE(2) correction in closed form, and report a measured covariance. The learned counterpart to camera-map-localization — written to be read, with a roadmap to distillation, pruning, INT8 and TensorRT.

Argoverse 2-style trajectory forecasting project comparing Linear, LSTM, Transformer, and Diffusion models.

Render, score, and compare Argoverse 2 driving scenes — annotated MP4s and per-city complexity graphs from a local Flask app.

HypoTrack-MF: evidence-gated cross-frame mode alignment for continuous motion forecasting

An LLM-powered learning pipeline for novel driving-scenario discovery from autonomous-driving sensor data.