I’ve been working on a project to figure out a way to achieve continuous movement in an urban environment captured by immersive panoramas. I came up with something that I call perspective scaling where frames between pairs of panoramas are recovered by applying transformations to them based on their depth maps.
The method runs entirely on the GPU and will yield a steady 60 FPS in web browsers on modern PCs. Behind the link is a demonstration of this method on a synthetic scene. The code is available on GitHub.
There are several issues that need to be worked out such as the disocclusion of static objects, the presence of moving objects in the scene, and transitions between panoramas need to be smoothed. I believe that a satisfactory solution can be found to all of these issues. Also, the current implementation of the algorithm is very wasteful which might be a problem on higher resolutions and mobile devices. I haven’t tested those cases.
Thanks for pointing this out. Blocking tor exit nodes was unintentional. I set up Cloudflare to save bandwidth as the panoramas in the demo make up close to a 100 MBs. A quick search did not yield a solution. Do you know how to fix this?
3 comments
[ 2.5 ms ] story [ 21.4 ms ] threadI’ve been working on a project to figure out a way to achieve continuous movement in an urban environment captured by immersive panoramas. I came up with something that I call perspective scaling where frames between pairs of panoramas are recovered by applying transformations to them based on their depth maps.
The method runs entirely on the GPU and will yield a steady 60 FPS in web browsers on modern PCs. Behind the link is a demonstration of this method on a synthetic scene. The code is available on GitHub.
There are several issues that need to be worked out such as the disocclusion of static objects, the presence of moving objects in the scene, and transitions between panoramas need to be smoothed. I believe that a satisfactory solution can be found to all of these issues. Also, the current implementation of the algorithm is very wasteful which might be a problem on higher resolutions and mobile devices. I haven’t tested those cases.
All feedback is appreciated.