Gaussian Splatting instructions using Postshot

Written By Eric Pare

I’m writing this article for my XangleCS users, meaning that this is optimized for multi-camera rigs, either using bullet-time or photogrammetry systems.

So, all this time, we were actually preparing the raw material to produce real 3D photos? For years, I’ve experienced a deep sense of dissatisfaction with the output from my bullet-time shots. I always knew there was more potential, and I got close with frame interpolation, but it never reached the smoothness we can now achieve with Gaussian Splatting. In this post, I'll walk you through my post-production workflow, specifically for Xangle Camera Server users like you.

This works with most of your bullet-time shots in Freeze mode (don’t try that with interval ones… we need to see the same moment from all angles). This will fix bad calibration and flickering, but it will fail if you’re trying to feed the machine with poorly synced images. Discard any smudgy, or out-of-focus images.

Required software

Post processing steps:

1- Locate your original files from your Xangle output folder c:\users\%userprofile%\xanglecs\output

2- (Optional, for better quality) Open your raw files in Lightroom. Apply AI denoise, color grading, and tweak Texture and Clarity a bit before exporting to PNG 16-bit. Skip this step if you're shooting in JPG.

3- Run Reality Capture and add your input files (Workflow / Inputs).
If you used any standard markers, start by automatically detecting these.
Click Alignment / Align Images, and make sure all of your cameras are properly registered

4- Export Camera Poses:

  • Navigate to the ALIGNMENT tab and select Registration under the Export sub-tab.

  • In the Export Registration File dialog, set Save as type to Internal/External camera parameters to save poses as a .csv file.

  • In the Exporter Settings window, use the default settings.

5- Export Point Cloud:

  • Under the Export sub-tab, select Point Cloud.

  • In the Export Point Cloud dialog, set Save as type to Sparse point cloud as Polygon File Format (*.ply).

  • In the Exporter Settings window, use the defaults, but adjust:

    • Export vertex colors: Set to True.

    • Export ascii: Set to False.

💡 Tip: Save both the .csv and .ply files in the same folder as the images. Drag and drop the folder into Postshot to import the entire dataset easily.

6- Run Postshot. Drop your PNG or JPG files (Postshot works better with "badly calibrated" images) along with the CSV and PLY files generated by Reality Capture.

7- Once the training is finished, create a Camera and select it from the dropdown menu as shown below.

8- Navigate around your subject, add key points, and have fun exploring your footage with the virtual camera. You can find a complete tutorial on Postshot key points here:

9- Finally, export your MP4 file from Camera/Output with the following settings:

Sample output:

Eric Pare
Next

Canon R100 tests - bullet-time & photogrammetry