Hello my friends on the internet. If you love 3D gaussian Splatting, stick around and get hypnotized by splats and floaters!
This experiment is to test how a dense point cloud affect 3D gaussian Splat training.
The idea is to use a dense point cloud to assist in converging splats faster. Will it work or will it be a huge waste of time? Let's find out my friends! Here we go!
The left side is a dense point cloud and right side is sparse.
Here is the training in realtime. You can see. The dense point cloud side converges faster. It is approximately one splat per point. The sparse side is blurrier. Look at the text Philtrust Bank and the pillar details!
This took less than 30 seconds. Normally without a dense point cloud, details starts to be clear much later depending on the scene.
Further training, will eventually even out both sides .
Splats, splats, splats are coming. Hey this is just a test. Chillax.. We gonna delete them later ok?
Hope you enjoy this experiment and see you next time!
Software: Adobe Lightroom, XNView, Reality Capture, Metashape Pro, Postshot
Edit: Davinci Resolve
Music: Suno.ai
26 сен 2024