RozDavid / LanguageGroundedSemseg

Implementation for ECCV 2022 paper Language-Grounded Indoor 3D Semantic Segmentation in the Wild
98 stars 14 forks source link

How can I get the predicted result? #14

Closed MurphyJUAN closed 1 year ago

MurphyJUAN commented 1 year ago

Hi, I am using ScanNet3D200 to do the semantic segmentation. If I want to get and export the predicted result to .txt file, how can I do that? (Should I use the 'save_predictions()' function? but how?) Thank you.

RozDavid commented 1 year ago

Hey @MurphyJUAN,

I assume you would like to submit it to the benchmark, where instead of the voxelized coordinates, you would rather want to evaluate on the full cloud. For this you should use first the standard evaluation pipeline followed by the test_pointcloud method which is automatically called if you set the test_original_pointcloud flag to true.

With that you associate the point to voxel predictions with a simple KNN search, that you can save with a simple numpy save command in either npy or txt format. The saving you can do with either the save_predictions function, or you can add an etxtra line simply to test_pointcloud.

Hope this helps, David

MurphyJUAN commented 1 year ago

I see it. Thank you so much for your reply. 🤩