Unsupervised learning on point clouds from driving data

[Waleed15]

Hello, I want to train the Voxel-MAE on my own dataset and extract the features to cluster the point clouds in an unsupervised manner. It would be helpful if you could suggest if is it possible with Voxel-MAE, if yes please give a brief overview of how to utilize your implementation for this case.

[chaytonmin]

Hello, I want to train the Voxel-MAE on my own dataset and extract the features to cluster the point clouds in an unsupervised manner. It would be helpful if you could suggest if is it possible with Voxel-MAE, if yes please give a brief overview of how to utilize your implementation for this case.

You can have a try. First pre-train Voxel-MAE, then extract the features from features from https://github.com/chaytonmin/Voxel-MAE/blob/459fafc67300a63a345edeb4701cd2adf24e4838/pcdet/models/backbones_3d/voxel_mae.py#L212.

I do not know whether it works. Maybe you can try some clustering-based self-supervised learning methods, such as SwAV, DeepCluster.