Idea for improving the Gaussian Process

[deebuls]

As mentioned before the problem with GP method is that you are not passing any information from the old frame to the next frame. nge

The solution :

1. save the previous frames latent space
2. in the next frame use the camera pose to calculate the distance and give it to the GP and propagate the previous latent space
3. No fuse the propagated latent space with the current latent space, Use addition to fuse

This will ensure 2 things

1. It propagates the previous frame information to the current frame
2. The distance vector and the GP makes sure how much old latent needs to be added .

Take a look . I think this is more important than the transformer because this will be a good story to say why GP failed and how we fixed it .