I'm currently working on a 3D object detection project using PointCloud data captured from a Realsense D405 camera. Here's my current setup:
- I've collected custom datasets from a Realsense D405 camera and formatted them to match the SUNRGBD dataset structure
- I'm using the TR3D model () for detecting 9 different objects
- However, I'm not satisfied with the detection performance I'm getting with TR3D
I'm specifically looking for:
- Models that rely solely on PointCloud data (x,y,z,r,g,b) without leveraging 2D image features
- Ways to improve TR3D's performance
- SOTA models that work well with SUNRGBD dataset format
- Any recommended models that can be trained with custom PointCloud datasets
I've searched through Papers With Code and GitHub but haven't found suitable open-source alternatives yet. Any suggestions or guidance would be greatly appreciated.
Development Environment:
- Ubuntu 22.04
- ROS2 Humble
- Python & C++
发布者：admin，转转请注明出处：http://www.yc00.com/questions/1744361793a4570506.html
评论列表（0条）