KIO Outdoor Test — 25.09.2025
Hardware:
• Camera: DJI Osmo Action 4 ×1
Software:
• Training Pipeline: Jawset Postshot 1.0 , Splat 3
Workflow:
The path was walked once in a zig-zag pattern at chest height, followed by a walk around the sorting container.
The purpose of this scan was to test splat quality, not to capture the entire space.
• Height: ~1.5 m
Dataset:
• Total number of images: 420
• Date of Capture: 25.09.2025
• Time to Capture: 3 Minutes
• Operator: A.B.
Key Observations:
Abandoning the 360-camera approach in favor of an action camera shows a lot of potential.
Pros:
• No need for a virtual camera rig in Blender, reducing potential points of failure in the pipeline.
• Importing the dataset directly into the training step preserves the native 4K resolution per image, compared to converting 360° footage into equirectangular images and re-rendering them in Blender.
Cons:
• The capture path requires more planning compared to the 360° workflow.
• longer capture time.
KreativInstitut.OWL
@kreativinstitutowl