Semantic segmentation : How to generate my own room_filelist.txt about pointnet

[themmes]

Could you maybe explain your data structure. Does it have a spatial partitioning comparable to rooms at all? If so, do you know what HDF5 batches correspond to those spatial partitions? If so, you could generate a room_filelist for your specific case.

[Vonisoa]

Thanks for your answer, I use a structure like in Stanford3dDataset_v1.2_Aligned_Version_3 (Area_1/conference_room_1/Annotations/), but I don't know how HDF5 partitions my dataset and even the partition by HDF5 in Stanford3dDataset_v1.2_Aligned_Version_3 is unclear fo me.

[themmes]

@Vonisoa Did you close the topic because you found a solution for your problem? If so, could you share your findings?

[Vonisoa]

Sorry, I accidentaly closed it.

[Vonisoa]

Btw, I run the gen_indoor3d_h5.py again to get it but still want to know how to generate the room_filelist.txt.

[charlesq34]

Hi @Vonisoa,

The filelist is written by code here:

pointnet/sem_seg/gen_indoor3d_h5.py

Line 80 in 4afd46d

fout_room.write(os.path.basename(data_label_filename)[0:-4]+'\n')

I think you only need to comment out the insert function below it, so that it will skip the data extraction (76-78) and insertion (82-83). The order should be deterministic for the filelist across runs.

hope it helps,
Charles

[Vonisoa]

Thank you very much, it really helps!