Do you have any suggestions for creating high quality photogrammetry models?

Creating a good photogrammetry model involves careful planning and capturing. Here are some key considerations (items highlighted in orange are most relevant to Nira):

1. Capture Quality

• Consistent Lighting – Avoid harsh shadows or overexposure. Overcast or diffused lighting works best.  This is not something you can always control, but if you can capture in the early morning while there is an overcast, that is preferred.
• Overlap – Ensure 60-80% overlap between photos to provide sufficient coverage.
• Angles and Coverage – Capture the subject from multiple angles (top-down, side, oblique). For objects, perform full 360-degree passes at different height levels.
• Resolution – Use a high-resolution camera to capture finer details. Using a lower resolution camera means you'll need to take more photos and move closer to the subject, which isn't always feasible (e.g., cell tower antenna interference may prevent your drone from having close proximity).
• Distance to Subject – Maintain a consistent distance to avoid scale distortion.
• Focus – Ensure sharp images by locking focus and using a narrow aperture for greater depth of field.
• RAW - This is not always recommended, but we sometimes suggest capturing in RAW to have greater control over the photos during post-processing.  While models are typically reconstructed using JPG images, starting with RAW files provides more flexibility if adjustments are needed. However, RAW files are larger and introduce an additional step in the workflow.  So this approach is only recommended when time and budget allow for producing the highest quality models but may be unnecessary for most standard captures.

2. Camera Settings

• ISO – Keep ISO low to minimize noise.
• Aperture – Use f/8 to f/11 for balanced sharpness and depth.
• Shutter Speed – Adjust based on lighting to avoid motion blur.

3. Model Processing and exporting

• Alignment – Carefully align photos during processing to reduce errors or gaps. If the alignment step is not executed properly, it will negatively impact all subsequent stages.
• Mesh Simplification – Decimate or simplify the model for flat surfaces to avoid excess triangles.  Most photogrammetry tools generate more triangle mesh data than necessary. For flat surfaces like building facades, simplifying or decimating the model is recommended. Typically, 50 to 100 million triangles is more than enough for most captures.  
• Clean-Up – Remove unwanted artifacts and refine edges.
• Texture Resolution – Use the highest possible texture resolution to preserve detail.  Many photogrammetry tools allow you to set the texture resolution, such as 4K, 8K, 16K, or up to 32K. Use the highest resolution available (e.g., 16K or 32K) and as many textures as required is ideal.  A single 32K texture (one billion pixels) contains the same pixel data as sixteen 8K textures (also one billion pixels). Therefore, if you need 256 x 8K textures (4 gigapixels), creating 10 x 32K textures (10 gigapixels) is preferable.  This reduces the file count, while preserving the same level of detail, and ultimately improves loading.
• Export in Parts – Export large models as separate parts to improve manageability and performance.  Some photogrammetry tools, such as RealityCapture, Bentley iTwin, and DJI Terra, default to or offer the option to export models as parts.  However, others like 3DF Zephyr and Metashape export the model as a single object, which is less ideal.

By focusing on these fundamentals, you can consistently produce high quality photogrammetry models suitable for various applications.