CatFree3D: Category-agnostic 3D Object Detection with Diffusion
This addresses generalization issues in 3D object detection for autonomous vehicles and robotics, though it appears incremental as it builds on existing diffusion and decoupling concepts.
The paper tackles the problem of limited generalization in image-based 3D object detection by introducing a diffusion-based pipeline that decouples 3D detection from 2D detection and depth prediction, achieving state-of-the-art accuracy and strong generalization across categories and datasets.
Image-based 3D object detection is widely employed in applications such as autonomous vehicles and robotics, yet current systems struggle with generalisation due to complex problem setup and limited training data. We introduce a novel pipeline that decouples 3D detection from 2D detection and depth prediction, using a diffusion-based approach to improve accuracy and support category-agnostic detection. Additionally, we introduce the Normalised Hungarian Distance (NHD) metric for an accurate evaluation of 3D detection results, addressing the limitations of traditional IoU and GIoU metrics. Experimental results demonstrate that our method achieves state-of-the-art accuracy and strong generalisation across various object categories and datasets.