LEF: Late-to-Early Temporal Fusion for LiDAR 3D Object Detection

TL;DR

This paper introduces LEF, a novel late-to-early recurrent feature fusion approach for LiDAR-based 3D object detection, improving shape and pose recognition by integrating temporal features more effectively.

Contribution

The paper presents a new late-to-early recurrent fusion scheme with window-based attention and a stochastic FrameDrop training method, enhancing detection accuracy especially for large objects.

Findings

Improved detection performance on Waymo dataset.

10× reduction in fused sparse features using segmentation.

Enhanced recognition of challenging large objects.

Abstract

We propose a late-to-early recurrent feature fusion scheme for 3D object detection using temporal LiDAR point clouds. Our main motivation is fusing object-aware latent embeddings into the early stages of a 3D object detector. This feature fusion strategy enables the model to better capture the shapes and poses for challenging objects, compared with learning from raw points directly. Our method conducts late-to-early feature fusion in a recurrent manner. This is achieved by enforcing window-based attention blocks upon temporally calibrated and aligned sparse pillar tokens. Leveraging bird's eye view foreground pillar segmentation, we reduce the number of sparse history features that our model needs to fuse into its current frame by 10$\times$. We also propose a stochastic-length FrameDrop training technique, which generalizes the model to variable frame lengths at inference for improved performance without retraining. We evaluate our method on the widely adopted Waymo Open Dataset and demonstrate improvement on 3D object detection against the baseline model, especially for the challenging category of large objects.