2024年4月16日 · 本文提出了一种名为 VSRD（Volumetric Silhouette Rendering for Detection）的新型弱监督 3D 检测框架。 这种训练框架不需要使用任何3D GT标签或者LiDAR点云，只凭借多视角的2D信息来训练3D检测器。

2024年3月29日 · To tackle this problem, we propose a novel weakly supervised 3D object detection framework named VSRD (Volumetric Silhouette Rendering for Detection) to train 3D object detectors without any 3D supervision but only weak 2D supervision.

VSRD optimizes the 3D bounding boxes and residual signed distance fields (RDF) for each target frame. The optimized 3D bounding boxes can be used as pseudo labels for training of any 3D object detectors. Sampled target frames in each sequence are split and distributed across multiple processes, each of which processes the chunk independently.

detection framework named VSRD (Volumetric Silhouette Rendering for Detection) to train 3D object detectors with-out any 3D supervision but only weak 2D supervision. VSRD consists of multi-view 3D auto-labeling and subse-quent training of monocular 3D object detectors using the pseudo labels generated in the auto-labeling stage. In the

2024年4月9日 · 为了解决这个问题，我们提出了一种新颖的弱监督三维目标检测框架，名为 VSRD（Volumetric Silhouette Rendering for Detection），用于在仅具备弱二维监督而无三维监督的情况下训练三维目标检测器。

In this paper, we propose a novel weakly supervised 3D object detection framework named VSRD, which consists of multi-view 3D auto-labeling and subsequent training of monocular 3D object detectors using the pseudo labels generated in the auto-labeling stage.

Duties included: As a researcher intern, we developed VSRD, a novel method for monocular 3D objectdetection with weak 2D supervision, avoiding the need for3D labels. Our approach utilizes multi-view 3D auto-labeling to generate pseudo labels for training.

我们提出了一种新的弱监督三维物体检测框架VSRD，利用自动标注阶段生成的伪标签在多视图三维自动标注和单目三维物体检测训练中优化三维边界框，实验表明我们的方法优于现有弱监督三维物体检测方法。

A new method called VSRD helps detect 3D objects in autonomous driving using only 2D supervision instead of expensive 3D labels. It uses advanced rendering techniques to train detectors effectively. The approach outperforms other weakly supervised methods on …

2024年4月24日 · 深度信息通过两种主要设计来利用。首先，我们引入一个提取深度特征FD的深度头D，利用深度特征 FD 来生成深度图 Dp，其中深度图的生成由深度图 Dgt 的伪地面实况监督，深度图 Dgt 是由现成的深度估计器通过使用焦点损失 [21] 来预测的深度损失 Ldep。