about the mean and std of part

[cocarecycler]

hi，thanks for your work

In the code, I see the mean and std of the upper part. I want to know how this is calculated.

self.register_buffer('mean_upper', torch.tensor([0.1216, 0.2488, 0.2967, 0.5027, 0.4053, 0.4100, 0.5703, 0.4030, 0.4078, 0.1994, 0.1992, 0.0661, 0.0639], dtype=torch.float32))

[exitudio]

Hi, You can ignore that part. I was just experimenting with a separated codebook (upper and lower), but I ended up not using it in the paper. Everything under is_upperlower is unused.

[Heibaiii]

Hi, You can ignore that part. I was just experimenting with a separated codebook (upper and lower), but I ended up not using it in the paper. Everything under is_upperlower is unused.

hi, a separated codebook (upper and lower) is a normal method in human motion works, have you tested it? Is the result better or worse after adding the separated codebook ? Thanks for your work!

[cocarecycler]

Hi, You can ignore that part. I was just experimenting with a separated codebook (upper and lower), but I ended up not using it in the paper. Everything under is_upperlower is unused.

THANKS

[cocarecycler]

Hi, You can ignore that part. I was just experimenting with a separated codebook (upper and lower), but I ended up not using it in the paper. Everything under is_upperlower is unused.

hi, a separated codebook (upper and lower) is a normal method in human motion works, have you tested it? Is the result better or worse after adding the separated codebook ? Thanks for your work!

it looks the same

[exitudio]

Hi, You can ignore that part. I was just experimenting with a separated codebook (upper and lower), but I ended up not using it in the paper. Everything under is_upperlower is unused.

hi, a separated codebook (upper and lower) is a normal method in human motion works, have you tested it? Is the result better or worse after adding the separated codebook ? Thanks for your work!

In BAMM, I just tried to inpaint body part following my previous paper, MMM. In MMM, I found that using separate codebooks led to worse performance.

P.S. 1: But MoGenTS shows that separate codebooks can improve performance. P.S. 2: However, my main focus was to overcome body part inpainting for Masked Motion Models. Which I solved this without relying on separate codebooks in ControlMM. (The code will be released soon.)