Ortho not rendering full selection area

I’m encountering unexpected behavior with orthos, only rendering half of what is obviously there in the textured model. Ortho1 shows in perspective view the fully textured model of a bridge scene in perspective looking down on the deck, with the orthoprojection region and direction indicated. Ortho 2 shows the result, half missing. Ortho 3 shows another orthoprojection looking up into the underbelly of the same bridge, this to isolate away from questions about system resources not handling the load, 232 8K tiles. Any thoughts what’s causing this half done business? Thanks.



Hello @BenjvC
Are the orthos created from the same reconstruction region?
Can you try to create a new reconstruction region and then set an ortho on that new region?
Is this also happening when you’ll close and reopen the project?
And can you try to move the bottom part of the region a little bit lower?

Also, I suppose it is not necessary to use such small ortho pixel size.

Thanks. I tried same rec. region and separate, tried lowering the lower plane, closing project, new project, and restarting PC, nada. But then lowering the pixel density, that finally worked! Why? thing is, I need full resolution, the whole point of the exercise being hairline crack detection.

Not to forget, the same size region for the ortho looking up into the underbelly was able to render at full resolution. The model has 232 8K tiles, which is a lot, but my PC didn’t sweat it, don’t believe it’s system resources. I just tried rendering the missing upper half, and that worked. Interesting, but how to make sense of whatever wall I hit. Again, I have 196 GB RAM, 24 GB video RAM, not seeing the pressure, is this a bug?

Oops, may have just answered my own question. While saving this last ortho, I got a Out of disk space error, most likely why I couldn’t render the whole region. I’ll confirm.

In RealityCapture there is a restriction for ortho creation. Or, for the export of the ortho.
Limit for non-tiff options: both sides of a projection should be shorter than 40000 pixels. The JPEG images are not limited only resolution-wise, but the size of a file can also be limited. I was not able to find a specific limitation, but the highest value I found was 50MB which would not be sufficient for a 50000 x 36000 image. This is just figurative, but a 2-megapixel camera produces a 6 MB image. The 50000 x 36000 image has 1800 megapixels, which is 900 times more. That means that it would have at least 5.4 GB which is a lot for a JPEG.
To export large ortho projection a TIFF format, or in this case, a BigTIFF format should be used (possible when exporting TIFF).

As your project is not georeferenced, you don’t know the exact value for the pixel size. For your case it could be smaller than needed (as there are only general units). The estimated size is not always the best one (I suppose it is a lot of times lower as the GSD of the model, which is not necessary to go lower of that value).

I suppose it is not a bug, but system restrictions.

What do you mean by “separate”? Like resized region?

What is the camera focal length, distance from the captured object and the sensor’s size?

42 MP, 21mm prime, full-frame sensor, about 3’ distance to subject.

If as you suggest this inability to render the full selection area stems from limitations unrelated to RC, how do you reconcile my being able to render at that same resolution when looking up into the belly of the bridge? I do see that some of the section in the underbelly, e.g. the rec. region clipping the base of the girders, that pixel real estate must be subtracted from the whole, which then might imply not reaching X threshold in play.

I suppose it is also RealityCapture’s restriction. The bottom ortho is slightly smaller. Maybe that it the border between rendered/not rendered.
Also, you won’t be able to open such big ortho almost everywhere. So it would be a good idea to create more smaller orthos with a wanted resolution.