DETR3D: 3D Object Detection from Multi-view Images via 3D-to-2D Queries

TL;DR

DETR3D introduces a novel 3D object detection framework from multi-view images that directly manipulates 3D predictions, outperforming existing methods in accuracy and inference speed on the nuScenes benchmark.

Contribution

It proposes a top-down 3D detection approach using sparse 3D queries and multi-view features, eliminating the need for depth prediction and post-processing.

Findings

Outperforms previous methods on nuScenes benchmark

No need for post-processing like non-maximum suppression

Faster inference speed due to direct 3D prediction

Abstract

We introduce a framework for multi-camera 3D object detection. In contrast to existing works, which estimate 3D bounding boxes directly from monocular images or use depth prediction networks to generate input for 3D object detection from 2D information, our method manipulates predictions directly in 3D space. Our architecture extracts 2D features from multiple camera images and then uses a sparse set of 3D object queries to index into these 2D features, linking 3D positions to multi-view images using camera transformation matrices. Finally, our model makes a bounding box prediction per object query, using a set-to-set loss to measure the discrepancy between the ground-truth and the prediction. This top-down approach outperforms its bottom-up counterpart in which object bounding box prediction follows per-pixel depth estimation, since it does not suffer from the compounding error introduced by a depth prediction model. Moreover, our method does not require post-processing such as non-maximum suppression, dramatically improving inference speed. We achieve state-of-the-art performance on the nuScenes autonomous driving benchmark.