Why do you choose to use blender to render condition images? about champ

[Leoooo333]

Hi @hangg7! We use Blender to render all conditions.

The pyrender in HMR2 is actually to visualize the overlay semantic map onto the reference image so that users can check and adjust its SMPL's shape parameters in --figure_scale. For example, given an extremely fat man image, the SMPL predicted from 4D-Humans may not perfectly match his figure. When running the SMPL Inference script, the pyrender will render the semantic map and save the overlay results under reference_imgs_folder/visualized_imgs, where users can check and then manually adjust --figure_scale if needed.

Since there are some engineering problems inside SMPL & Rendering part, we haven't PR it now. We are trying to fix it and aim to release it within this week.

[hangg7]

Thank you for your quick reply! Now if we want to render conditions without blender, is it possible to do it, as long as we align with your blender script, or is there something we should keep in mind other than aligning the rendering logic (e.g. how do you set cameras and things like that)?

[Leoooo333]

@hangg7 Yeah I think so. All of the renderer settings (including the camera) are in that blender file and rendering script. It's the same camera setting with 4D-Humans, with the camera fixed at the origin and facing to +Y axis. By the way, seems like you have a faster renderer. We are welcome to contribute!👏

[hangg7]

Would love to contribute what I can. I am looking at your own branch and maybe I can submit some PR there. Closing this!