Training with own data

[ghost]

Hi, thank you for giving me new insights into my research.

I'm trying to train with my own point cloud data (with only 1 Channel not RGB) but I couldn't find where should be modified.

Actually, I don't have scannet data right now, So I can't survey structure of scannet data.

So, Could you let me know proper form of data (label) structure and where should be modified in the codes to train only 1 channel data?

[llijiang]

Hi,

The data loaded from the file for a scene [here] is of the form:

• xyz: (N, 3), point coordinates
• rgb: (N, 3), rgb values normalized to (-1, 1)
• label: (N), semantic labels ranging from 0 to nClass - 1, unannotated points are labeled with -100
• instance_label: (N), instance labels ranging from 0 to nInstance - 1, unannotated points are labeled with -100

To fit into your own data, some possible positions to be modified are:

1. Load your 1-channel data [here] instead of rgb, and modify the codes related to the dimension (3 -> 1) in the data loader.
2. Change the input_channel [here] to 1.
3. Some other configs (e.g. cluster_radius) may need to be adjusted according to your data.

Hope it helps.

[heesooy]

Hi @llijiang,

I read in your paper that you tested the model on the S3DIS dataset. Do you have a dataloader and config file for S3DIS that you could provide, and suggestions for any other parameters to change? I likewise can't access the ScanNet data and I'm unsure how to adapt the code since I don't know how the ScanNet data is structured.

[PatrickFeng]

Hi @llijiang,

I read in your paper that you tested the model on the S3DIS dataset. Do you have a dataloader and config file for S3DIS that you could provide, and suggestions for any other parameters to change? I likewise can't access the ScanNet data and I'm unsure how to adapt the code since I don't know how the ScanNet data is structured.

@heesoo4199 You don't need to download all the data from ScanNet. Only 4 types of data, _vh_clean_2.ply, _vh_clean_2.labels.ply, _vh_clean_2.0.010000.segs.json and .aggregation.json, are required and could be downloaded with command python download-scannet.py -o [path_to_save_scannet] --type [type_of_data]. scannetv2-labels.combined.tsv would also be downloaded along with the command.

[BanaBear]

@PatrickFeng hello, i would like to know where the download-scannet.py is.

[PatrickFeng]

@PatrickFeng hello, i would like to know where the download-scannet.py is.

@BanaBear You can refer to https://github.com/ScanNet/ScanNet. ScanNet team will reply after you sending the ScanNet Term of Use mail to them. You could get the tutorial on how to download the dataset in that mail.

[lifeiwen]

@PatrickFeng hello, i would like to know how to prepare my owm data, I have a batch of point cloud data sets with only xyz coordinates, How to construct the type of data set for training？thank you for your reply

[tabarkarajab]

can you kindly tell me from where you converted your .ply dataset into seg,json formats?