The idea is to detect object point cloud(chair, desk …etc) from the original point cloud. Further, using it in the construction industry. E.g. detecting artifacts such as pipes, workers, cranes …etc.
The paper and code are originally from Facebookresearch/votnet. I was using SCANNET dataset as an example and verified its accuracy. Below are the steps to reproduce the experiment.
Firstly, setting up pytorch environment with GPU support. Then download the dataset through official permission. Then I found it was not possible to train the model as my GPU has limit RAM. I still need to fix my hardware issue at some point. Currently, I can only use pre-trained model. Thus I tested one point cloud with a solo desk and the model showed a good detection. See pic below,
