Visualize Tensor image without copying out of gpu?

I want to visualize tensor images generated in the gpu with pytorch with as little delay as possible and minimize usage of system resources. Is there a solution that would involve torch3d and a 3D rendering context that pytorch could feed tensor images into as textures? I’m not sure if torch3d has the ability to directly render like that.

(copied from here
Display a pytorch gpu tensor image without copying to cpu memory? - Stack Overflow)