Reconstruct Anything

Convert Raw RGB-D to tree-structure scene(maybe in unity), for more

1. Raw point cloud (voxel) to semantic segmented cloud
2. classify the segmented object into different structure and establish parent-child relationship
3. identify the state of object
   1. pose like, numbers (\\)
   2. state like, open/close (LLM)

发现和lff近期发表的一篇文章思想非常一致 https://arxiv.org/html/2410.07408v1

和场景理解的对比