Suppose we have access to the actual camera intrinsics, will it improve performance if we just replace the K with our recorded camera intrinsics?
The K in the demo.py code for example.
Yes, of course! If you have access to your image's focal length, you should use that in the demo. The demo was built so that users can run it on any image, e.g. images from COCO.
Suppose we have access to the actual camera intrinsics, will it improve performance if we just replace the K with our recorded camera intrinsics? The K in the demo.py code for example.