hello! Thank you for sharing your wonderful research.
I am currently implementing a room stage. I think you optimized the cube's transformation matrix by creating a cube mesh and rendering the cube's depth and mask.
The question is, when calculating the mask loss in room optimization in your paper, it appears that IOU was obtained separately for the wall, floor, and ceiling. How did you render all the masks with one cube?
Or did you just create a mask for the entire cube and calculate the loss for the entire wall, floor, and ceiling?
Another thing I'm curious about is how much space was allocated behind the camera when initializing the room (initializing the cube)? In the implementation I'm currently trying, when initializing the cube with a point cloud, it seems that only a small amount of space is allocated behind the camera.
The first problem has been solved. This was easily solved by creating a mesh for each of the wall, floor, and ceiling and optimizing the same transformation matrix for all of them.
hello! Thank you for sharing your wonderful research.
I am currently implementing a room stage. I think you optimized the cube's transformation matrix by creating a cube mesh and rendering the cube's depth and mask.
The question is, when calculating the mask loss in room optimization in your paper, it appears that IOU was obtained separately for the wall, floor, and ceiling. How did you render all the masks with one cube?![image](https://github.com/andyyankai/psdr-room/assets/24850969/be929072-ab83-4248-8858-2387e7f7f061)
Or did you just create a mask for the entire cube and calculate the loss for the entire wall, floor, and ceiling?
Another thing I'm curious about is how much space was allocated behind the camera when initializing the room (initializing the cube)? In the implementation I'm currently trying, when initializing the cube with a point cloud, it seems that only a small amount of space is allocated behind the camera.