One reason I made this is because I think Gaussian splats are the future and I think we can be embedding these into all kinds of immersive experiences. I think newsrooms should be making and using these for storytelling even in same-day beat reporting.
February 21, 2025 at 4:23 AM
One reason I made this is because I think Gaussian splats are the future and I think we can be embedding these into all kinds of immersive experiences. I think newsrooms should be making and using these for storytelling even in same-day beat reporting.
The biggest challenge I had was trying to color match images taken on four different sensors at different times of day, which gsplat's bilateral grid feature does a very good job of handling. Pictured is an early failure from PostShot.
February 21, 2025 at 4:23 AM
The biggest challenge I had was trying to color match images taken on four different sensors at different times of day, which gsplat's bilateral grid feature does a very good job of handling. Pictured is an early failure from PostShot.
To build the model, @daylen.bsky.social and I flew drones around the tower and took thousands of files. I then somewhat laboriously aligned these images in RealityCapture
February 21, 2025 at 4:23 AM
To build the model, @daylen.bsky.social and I flew drones around the tower and took thousands of files. I then somewhat laboriously aligned these images in RealityCapture
I implemented a JS decoder for a recent compression scheme for splats called Self Organizing Gaussians (fraunhoferhhi.github.io/Self-Organiz...) that leverages classic image codecs. It brings the size of the scene down to under 30MB!
I implemented a JS decoder for a recent compression scheme for splats called Self Organizing Gaussians (fraunhoferhhi.github.io/Self-Organiz...) that leverages classic image codecs. It brings the size of the scene down to under 30MB!