Efficient Monocular 3D Object Detection through Teaching Assistant Knowledge Distillation
The core message of this paper is to propose a Teaching Assistant Knowledge Distillation (MonoTAKD) framework that enhances the efficiency of distilling 3D information from a LiDAR-based teacher model to a camera-based student model for monocular 3D object detection.
Enhancing Monocular 3D Object Detection with Complementary Depth Predictions
The core message of this paper is that existing monocular 3D object detection methods suffer from a coupling problem where multiple depth predictions tend to have the same error sign, limiting the accuracy of the combined depth. To address this, the authors propose to increase the complementarity of depths by introducing a new depth prediction branch that utilizes global depth clues and by exploiting the geometric relations between multiple depth clues to achieve complementarity in form.
Roadside Monocular 3D Object Detection Improved by Leveraging 2D Detection Prompts
The core message of this work is that exploiting a well-trained 2D object detector can significantly improve the performance of roadside monocular 3D object detection.
Improving Monocular 3D Detection of Large Objects with Segmentation in Bird's View and Dice Loss
Monocular 3D detectors struggle to generalize to large objects due to the noise sensitivity of depth regression losses. SeaBird, a novel pipeline, effectively integrates BEV segmentation supervised by the noise-robust dice loss to improve monocular 3D detection of large objects.
