Difference in Precision Between Parts and the Whole Object When Performing Photogrammetry on a Room

Here are two images.
One consists of an object created from 33 photos, while the other is an object within a room composed of around 900 photos.

Even though they contain the same photos, when only the objects were reconstructed, a higher quality is observed compared to the object within the fully reconstructed room.

Why does this phenomenon occur?

To maintain the same high quality as when reconstructing only the object, even when performing photogrammetry of the entire room, what should be done?

Here is the other one.

The default texture is a single 8k texture, this is spread quite thin on a whole room, try texturing using “fixed texel size” set to optimal. Then you need to make a render not take a screenshot, as the viewport will only render a single 16k texture and most likely the room will have a lot more textures than that

“Well, I’ve already used the ‘fixed texel size’ with the ‘optimal’ option. Are there any other methods?”

Did you also try making a render?

How many textures of what resolution is the model of the full room?

(8192x8192) x 192.

OK that will look terrible in the viewport. The viewport has a limitation of displaying a single 16k texture, which is equivalent to only 4 of your 8k textures, so you need to make a render to see the texture properly on your model.

However, when I compared the rendered results of the project created by inputting only 33 photos and the project created by inputting 1000 photos into the room, I noticed differences in the quality of the mesh and texture.

Is it possible that Reality Capture’s precision in finer details could slightly diminish as projects scale up?

Possibly, it really depends on the capture. But normally the quality is retained, but your render of the area should be significantly better than what you saw in the viewport, as when RC converts the 192 8k textures down to 1 16k texture, the texture you see in the viewport will be about 50 times worse than the texture that is actually on your model and the one that would be exported. The only thing that might make the quality of some parts worse is if you have photos from a lot further away that also “hit” the area, because they have far less precision they will most likely make the mesh and the texture worse, although both algorithms do account for this. you might be able to get away with disabling these images for meshing and texturing unless they are important for another area.

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.