Map-Det3D: Metric Feed-Forward 3D Reconstruction Prior for Multi-view 3D Object Detection

rsasaki0109 · x · 2026-08-18

Map-Det3D is a metric feed-forward 3D reconstruction prior for multi-view 3D object detection from streaming RGB inputs, accepted to ECCV 2026. The method maps a short temporal window into multiple views and repurposes a feed-forward metric 3D reconstruction model as its geometric backbone, predicting boxes directly in metric 3D space without the widely used 2D-to-3D lifting. Experiments show strong online performance and robust transfer without adaptation.

Original post →

More from Embodied

Embodied channel →