[ Volumetric Heatmaps Dimension ]

[IemProg]

Hi,

What's the dimensions of the of heatmaps before permute function in _downlayers block, Is it [Batch_size, Joints, Depth, Height, Width ] ?

if that's the case how can we apply permutation if number of joints is not the same as width, it will raise dimentionality mismatch given that you have applied permutation of:

mid_out = torch.cat([t.permute(0, 3, 2, 1) for t in mid_in.split(size, -3)], -3)

What does "C" letter refer to in Figure 5 of the paper, please ?

Thanks !

[anibali]

"c" is the number of channels for the intermediate features, which must be a multiple of the height/width. Note that the number of joints is not one of the dimensions (these are intermediate features, not heatmaps).

So the shape of mid_in is [batch_size, c, height, width] where height = width and c is a multiple of width.