Best practice for combining mapping images and supplemental manual shots for 3D mesh reconstruction?

[RSDSDivision]

Hi everyone,

I'm looking for some guidance regarding Drone2Map's 3D mesh reconstruction workflow when combining an automated mapping mission with supplemental manual shots.

For context, our organization officially uses Drone2Map 2026.1.0 and SiteScan.

For this particular project, I felt the 3D mesh output could still be improved, especially in terms of facade details and overall completeness. As an additional benchmark, I used the DJI Terra free trial to see how another reconstruction engine would handle the exact same dataset.

The dataset was acquired using a DJI Matrice 4E and consisted of:

• An Oblique Mission
• Additional manual/oblique shots captured around the subject to improve facade coverage and capture finer details

I then processed the same image dataset using default settings in Drone2Map, SiteScan, and DJI Terra.

I observed that DJI Terra appeared to leverage the supplemental manual/oblique shots more effectively, resulting in a more detailed and complete 3D mesh.

Before concluding that this is simply a difference between reconstruction engines, I'd like to ask if I'm missing a recommended Drone2Map workflow or processing configuration.

Some questions I have:

• Is there a recommended workflow when combining an Oblique Mission with supplemental manual shots?
• Is there a preferred processing template for this type of dataset?
• Are there any special considerations or best practices when incorporating supplemental manual/oblique shots into a Drone2Map 3D mesh workflow?

I'll attach side-by-side comparison images (Drone2Map vs SiteScan vs DJI Terra).

Important: All outputs were generated using the exact same image dataset. No additional images were added or removed between software packages.

I'm not trying to determine which software is "better." My goal is to understand whether there are recommended Drone2Map settings or best practices that would allow it to better leverage supplemental manual imagery.

Any advice, workflow recommendations, or examples from your own projects would be greatly appreciated.

Thank you.

[MarkBarker1]

Hi @RSDSDivision,

Thank you for reaching out, I am curious about what point cloud density processing setting you were using for the 3D outputs within drone2map (Options>General>Dense Matching)? To capture the very fine details we tend to recommend processing on the ultra point cloud setting as it will handle finer details better. Alternatively, we have a gaussian splats product if you have a capable GPU that can support its processing it tends to best the best option for detail.

Regarding your questions:

1. There is no specific additional workflow when combining the shots, but the general photogrammetry rules apply, such as ensuring the correct metadata information is present on the images and the camera view angle isn’t pointed past the horizon.
2. The recommended processing template would be the 3D products template. You can fine tune it after selecting the template to increase point cloud density or other settings such as output format types.
3. The same considerations as your first questions apply here. Drone2Map will attempt to adjust any input imagery to find neighboring matching images and derive tie points between them. So if your manual pictures cover the same subject matter then the photogrammetry engine will attempt to incorporate them in.

If you are willing to share your processing report I can review your settings. Additionally, if you are open to sharing your project imagery please let me know and I can look at it on my end to see if there are any areas of improvement I can suggest.

Thank you,

Mark

[RSDSDivision]

Hi Mark,

Thank you for your response and for offering to review the project.

I'd be more than happy to share the raw image dataset so you can process it on your end. Since the same images were used in Drone2Map, Site Scan, and DJI Terra, I think it would be a good opportunity to determine whether there are any Drone2Map settings or workflow adjustments that could improve the results.

Please let me know the best way to provide the dataset. It's approximately 493 images.

Thank you again for your assistance. I really appreciate you taking the time to look into this.

[MarkBarker1]

Hi @RSDSDivision,

Feel free to email me directly and I can get a shared folder going: [email protected]

Thank you,
Mark