drprojects / superpoint_transformer

Official PyTorch implementation of Superpoint Transformer introduced in [ICCV'23] "Efficient 3D Semantic Segmentation with Superpoint Transformer" and SuperCluster introduced in [3DV'24 Oral] "Scalable 3D Panoptic Segmentation As Superpoint Graph Clustering"
MIT License
601 stars 75 forks source link

[Environment Configuration] SPT and SPG Environment Compatibility #16

Closed pynsigrid closed 1 year ago

pynsigrid commented 1 year ago

Hi, thanks for your excellent work in point clustering. I am interested in this field and have been working on reproducing your previous work, the SuperPoint Transformer, alongside your previous excellent project, the SuperPoint Graph.

From my understanding, SPT includes point partitioning, which is similar to that found in SPG. However, I encountered some difficulties while attempting to install the required environment for SPG, while the installation process for SPT was much smoother and successful. Therefore, I am curious to know if the SPT environment also supports the SPG project, allowing me to conduct SPG under the SPT environment.

drprojects commented 1 year ago

Hi @pynsigrid , thanks for your interest in our superpoint-based projects !

As indicated on the SPG repo, this project is no longer maintained.

The current project does not support SPG. It is built so as to replace SPG and does everything better:

loicland commented 1 year ago

Author of SPG here. SPG and SPT use the same algorithm for partitioning: cut pursuit. All the boost functions of SPG have been reimplemented in SPT, so in theory, you could use the first level of a partitions of SPT to run SPG.

But the graph format are different, so there might be a bit of work to adapt it. And like @drprojects said above, SPT does all the same thing than SPG but better, so there is no real reason to use SPG anymore.

pynsigrid commented 1 year ago

Noted that. Thanks @drprojects @loicland for your kind explanation! It really helps me alot🥳