Question about Partition Implementation (P2, P3, P4) in Code

[JackWangPulse]

Thank you for sharing your open-source code!

While studying the partitioning strategy described in your paper, I noticed that the code implementation may not fully align with the description for P2 Partition (Distant joint, local frame), P3 Partition (Local joint, long frame), and P4 Partition (Distant joint, long frame).

For example, in the case of P2 Partition (N=2, K=2):

Example P2, N = 2, K = 2:

def type_2_partition(input, partition_size):  # partition_size = [N, K]
    B, C, T, V = input.shape
    partitions = input.view(B, C, T // partition_size[0], partition_size[0], partition_size[1], V // partition_size[1])
    partitions = partitions.permute(0, 2, 5, 3, 4, 1).contiguous().view(-1, partition_size[0], partition_size[1], C)
    return partitions

I’m concerned that this implementation might not produce spaced frames (as described in the paper) while still maintaining adjacent frames.

Could you please confirm if my understanding is correct or clarify the intended logic for this partitioning?

Any insights or suggestions would be greatly appreciated. Thank you in advance for your help! 😊

[drwangusm]

so many issues