Xiaoyang Wu

Thanks. I need some time to review the code (I need to download the dataset first and follow your code). By the way, I am going to publicize one work...

So, assume 8192 is sourced from 16 (seq_len) * 512 (num_points). As we already flatten the point cloud by merging batch_size and seq_len into on single concept of batch_size, how...

> That's great! I am excited to see how your experiments turn out and how these datasets contribute to the performance. > > By the way, I am adding more...

Change in_channel of model to 3

Check dtype of grid_coord, it should be int, not float

1. Adjust data.train(val, test).transform.collect.feat_key in config. 2. Change in_channels and out_channels in PTv3 model config

I am sorry for the long wait. I just finished downloading the raw Matterport3D dataset and will review your code this weekend! Thanks again for your contribution.

Hi, @hugoycj , would you mind contacting me with my email (xiaoyang.wu.cs@gmail)? As we recently introduced the Black formatter, we need to format, rebase, and force-push your code before merging....

> @SomnusKKK Hello, can I refer to your code? I am currently exploring the possibility of applying PTv3 to point cloud sequences. The point cloud sequence shape is (batchsize,Length,N,3). Check...

@hugoycj Sorry for the super super long wait. I finally got some time to handle this PR as I am working on scaling up self-supervised pre-training with more data. I...