View Selection Strategy not applied in published code?

[supersyz]

Thanks for you great work!
I notice that in the paper 8~12 view-points are heuristically selected for inference.
However, in the code, it seems to have only 6 fixed view-points:

self.candidate_camera_azims = [0, 90, 180, 270, 0, 180]
self.candidate_camera_elevs = [0, 0, 0, 0, 90, -90]
# AND 
selected_camera_elevs, selected_camera_azims, selected_view_weights = \
self.config.candidate_camera_elevs, self.config.candidate_camera_azims, self.config.candidate_view_weights

Is this the publised code a simplified version compared with the paper? Or do I misunderstand it?
Hope for your replay,
Best.

[rocksat]

same question here. I don't see this function in published code:

A area(· · · ) is a function that calculates the coverage area according to the given set of covering texels.

[qinyuxin010328]

Same question here, could not find the selection part

[qinyuxin010328]

Is it the code in/Hunyuan3D-2/hy3dgen/texgen/differentiable_renderer/mesh_render.py the selection part? It seems like it is enlarging the texture map untill the selection area is above 99% of the total area.
@torch.no_grad()
def fast_bake_texture(self, textures, cos_maps):

    channel = textures[0].shape[-1]
    texture_merge = torch.zeros(
        self.texture_size + (channel,)).to(self.device)
    trust_map_merge = torch.zeros(self.texture_size + (1,)).to(self.device)
    for texture, cos_map in zip(textures, cos_maps):
        view_sum = (cos_map > 0).sum()
        painted_sum = ((cos_map > 0) * (trust_map_merge > 0)).sum()
        if painted_sum / view_sum > 0.99:
            continue
        texture_merge += texture * cos_map
        trust_map_merge += cos_map
    texture_merge = texture_merge / torch.clamp(trust_map_merge, min=1E-8)

    return texture_merge, trust_map_merge > 1E-8`

`