Using batch point clouds as model input

[nghiemkythu]

Hi, thank you very much for your valuable research and for releasing the code.

I have a question about how to use the data transform and the model when the input is a batch of point cloud samples, instead of a single point cloud.

Currently, each sample in my dataset contains a point cloud with a different number of points. How i config the transform in "concerto.transform" and input to the model.

Any guidance or examples would be greatly appreciated.

Thank you!

[Gofinge]

Hey, please refer this: https://github.com/Pointcept/Concerto/blob/main/demo/3_batch_forward.py
Also point cloud are dynamic shape in Pointcept framework.

[nghiemkythu]

Thank you very much for your response. After obtaining the model output, is the following code the correct way to extract the features corresponding to each sample?

feat_batch = batch_out["feat"]
batch_ids = batch_out["batch"]

feat_sample = feat_batch[batch_ids == i]  # i is the index of the sample