Monocular 3D Object Detection
This technology is a novel end-to-end trainable Grouped Mathematical Differentiable NMS (GrooMed-NMS) for monocular 3D object detection.
Modern 3D object detection methods have immensely benefitted from the end-to-end learning idea. However, most of them use a post-processing algorithm called Non-Maximal Suppression (NMS) only during inference. Although there have been attempts to include NMS in the training pipeline for other tasks such as 2D object detection, they have been less widely adopted due to a non-mathematical expression of the NMS.
Description of Technology
This technology formulates NMS as a matrix operation and then groups and masks the boxes in an unsupervised manner to obtain a simple closed-forms expression of the NMS. GrooMeD-NMS addresses the mismatch between training and inference
pipelines and therefore forces the network to select the best box in a differentiable manner. GrooMeDNMS achieves state-of-the-art monocular 3D object detection results on the KITTI dataset performing comparable to monocular video methods.
- This technology enables a end to-end trainable closed-form mathematical differential Non-Maximal Suppression..
- GrooMeD-NMS addresses the mismatch between training and inference pipelines prevalent in monocular 3D object detection and therefore forces the network to select the best box in a differentiable manner.