Is there a bug in the calculation method of head_torso_alpha in the HTB-SR v2 model?

[Mayoiuta]

Thank you for the excellent code!
While studying the HTB-SR model, I noticed the following code in the file.
Here, the weights tensor is cloned:

Real3DPortrait/modules/real3d/super_resolution/sr_with_ref.py

Line 108 in a9d70c7

head_torso_alpha = weights_256.clone()

At this point, (head_torso_alpha > weights_256) should always be False.
Therefore, does the following line of code—

Real3DPortrait/modules/real3d/super_resolution/sr_with_ref.py

Line 109 in a9d70c7

head_torso_alpha[head_torso_alpha>weights_256] = weights_256[head_torso_alpha>weights_256]

—have no effect?

Real3DPortrait/modules/real3d/super_resolution/sr_with_ref.py

Lines 106 to 115 in a9d70c7

	elif hparams['htbsr_head_weight_fuse_mode'] == 'v2':
	# 用alpha-cat实现head torso的x的融合；替代了之前的直接alpha相加
	head_torso_alpha = weights_256.clone()
	head_torso_alpha[head_torso_alpha>weights_256] = weights_256[head_torso_alpha>weights_256]
	rgb = rgb * head_torso_alpha + rgb_torso * (1-head_torso_alpha) # get person img
	x = torch.cat([x * head_torso_alpha, x_torso * (1-head_torso_alpha)], dim=1)
	x = self.fuse_head_torso_convs(x)
	x, rgb = self.head_torso_block(x, rgb, ws, **block_kwargs)
	head_occlusion = head_torso_alpha.clone()