How to run PSSNet fully

[zwl995]

Hello, Dr. Gao, I want to run PSSNet completely. Are the steps to run the steps one after another:

1. Run the PSSNet_oversegmentation mode in step1 and only execute process_prediction
2. Readme execution according to step2

[WeixiaoGao]

Sorry for my late reply.

1. You can directly run` PSSNet_pipeline_for_GCN' to get all the required data for the next step.
2. For step-2, except for the Readme, please also check here.

[zwl995]

[zwl995]

[zwl995]

Is this the right way to organize the files? In addition, I placed the models provided by the SUMS warehouse under the models folder

[WeixiaoGao]

If you want to predict unseen data, that's the correct way to do it with a pre-trained model. However, you need planar and non-planar classification models for step 1 and another pre-trained model depends on the method in step 2.

[zwl995]

Thank you very much for your reply. My purpose is to classify models that I have never seen before.
I set operating_mode to "PSSNet_pipeline_for_GCN" according to your instructions, and I observed that it was executed in turn: Get_labels_for_plane_non_plane_from_semantic, Pipeline, PSSNet_oversegmentations, PSSNet_oversegmentations_evaluation, PSSNet_graph_construction, PSSNet_pcl_generation_for_GCN. However, after debugging during the training of Pipline, it was found that all seg_truth_train were-1, which caused errors in subsequent training.