Unable to Scale & Geo Reference

[Tokyo Warfare Project]

Hi.

I tried to place 3 control points, I took the coords from Google earth in UTM 52s and WGS84 projection, I know is not exat but at least I was expecting to get an aprox scalling.

That seemed to fail as the model was like 90º rotated. So I decided to go the lazy way as I just need it scaled and not geolocated.

I converted the ground control points to tie points and added a measure constraint.

When I add the ral distance between the control points the software says "The difference between the computed and the defined distance is over the limit" JA!! computed based on what!!! there is no way it can guess scale, and obviously does not, but the thing is that it does not allow me to force scale! I added 2 more distance constraints and same story.

Here a pic
https://1drv.ms/i/s!AuliwwO_siw1iDRDgCV595MiBjUV


I really whant to scale it beforehand I do the texturing because I think the texels size is messing up due to the small size of the model and the fixed texel size.

If I set it to have 100% quality the texels are GIGANTIC, and the quality is a turd.
https://1drv.ms/i/s!AuliwwO_siw1iDXOLJl6RQ0hnnva

For a texel size that would capture some decent detail, the quality percent is set to 1700% and that makes no sense.

Further, when I've a decently small texel size texture processing fails, bar goes red and shows no error msg.

I think this all is coming form the model having a ♥♥♥♥♥♥♥♥ of points with a fixed texel size but with an actual tiny metric size compared to real.

While till this stage the program performed like a beast, at this point it feels a bit broken.

I was able to export a 200M cloud tho, and in Cloud compare I expect I'll be able to export as an image and do the job of an orto export, quality is not bad but yet not as good as the real orto based on image should be.

That said, I've been trying to do the texturing in the simplified 2M poly model.

So, please help me with the scaling first, and lets see if the other stuff solves afterwards.


Edit. just goat a windows message of memory getting low and that it was gonna start clossing apps... I guess it is RAM issue but should a 16 gig ram be enought?. What a bout some virtual memory in a SSD. I checked c and has 80 gigs free. I was testing if dissabling many pics and taking only the top botton oriented ones would be enough to make the orto and bang. it crashed.

[zuduri]

Hello,

I also commented your previous post. To solve your scaling problem, have you tried to change all GCP to tie points? Then you can set the distance between tie points and click Update to scale your model to this distance.

[zuduri]

For the hardware requirements, please see this post:

https://support.capturingreality.com/hc/en-us/articles/115001524071-OS-and-hardware-requirements

[Tokyo Warfare Project]

Originally posted by zuduri:

Hello,

I also commented your previous post. To solve your scaling problem, have you tried to change all GCP to tie points? Then you can set the distance between tie points and click Update to scale your model to this distance.

Hi. as I told you
"I converted the ground control points to tie points and added a measure constraint."

But I'm gonna double check. and try recreating them fro zero. But yes, as I was understanding that otherwise the scale would be set due to three points having coordinates.

I've also thought of doing a relative point coordinates one being (0,0,0)
and the rest having relative coords to this point like (0,100,0) (20,10,0) instead of using UTM. this will expose if there is a bug or not

In regards to hardware requirements I've a 16gb ram i7 3820 and GTX 1080 so on the paper should be no problems.

In regards to memory it says:
- 2000 images (scans)/40K features per image - 16GB RAM

I'm running a 1200 img job and it peaks 16gb usage almost as quick as I hit the compute texture. but to be hones I think is related to the fact is says I'm using 1600% of texture quality, witch is happening because I'm using fixed texel size and the scale is messed up....

[Tokyo Warfare Project]

As you can see in this image, the control points are of type "TIE POINT" and error in distances is present
https://1drv.ms/i/s!AuliwwO_siw1iDY4fL8ODqGgqbY3

[Tokyo Warfare Project]

Good news!
The coordinates where projecting like in a globe sphere when using WGS84 UTM 52s. This is probably correct as it does place in the worldspace coords but is a bit ankward relative to the software grid.

If I use the coordinates with an euclidean projection they work as I did expect.

At leas I tested with the relative coordinates having point 0 in 0,0,0

x y z x y z
point 0 389544.01 --- |---3935086.89 --- |---0 --- |--- 0 --- |---0 --- |--- 0
point 1 389891.43 --- |---3935359.66 --- |---0 --- |---347.42 --- |--- 272.77 --- |--- 0
point 2 390475.55 --- |---3934307.51 --- |---0 --- |---931.54 --- |--- -779.38 --- |--- 0

[Tokyo Warfare Project]

Existing ortophoto seems to not update scalling.

You can see in the image that the 1000m graphic scale is almost the size of the distance constraint.
https://1drv.ms/i/s!AuliwwO_siw1iDx-9LcWKHUukL8A

You can see in the data panel that the distance contraint (measured by software value) is 441m

[Tokyo Warfare Project]

I can confirm that while distances seem ok, the model does not georeference.

Here you can see GCP and sitance errors (ok one has big error but others are acceptable to me)
https://1drv.ms/i/s!AuliwwO_siw1iD1hlsuL5Ed_HCGT

The components shows no georeference value, and texels do no adapt to scale values.
Component texturing optimal texel size is said to be 0.083989 meter per texel. But if I put that size it generates GIGANTIC texels, of like 60meters in model size.

I've yes clicked update button for scale changes to apply. The only way to georeference a component is to set coordinates beforehand? According to documentation update should scale all component elements.

I'm ruinng align process again.

[Tokyo Warfare Project]

If I re- run alignment then now it shows as georeferenced. PLEASE add option to update existing components to be georeferenced AFTERWARDS they've been aligned.

[Tokyo Warfare Project]

ok, as I was suspecting, the fixed texel size issues where due to model scale, and as model scale seems glitched if you try to scale after you align I couldnt get a decent output.

I re run alignment (draft) and re run meshing (preview) for a quick re-texturing at a proposed texel size. FINALLY the output seems DECENT!! Just coment that viewport resolution seems to not show full texture quality, further, I've seen to degrade from the non clip version to clip version. Here an snap from cliped mesh and rendered snapshot. I've mistekenly re-unwraped so I've not yet tested orto quality, I hope is as good as in the render.

https://1drv.ms/i/s!AuliwwO_siw1iD7DbTpMVx_8mILr

[zuduri]

When you use Update, it only scales and rotate existing component but the component remains not georeferenced. You have to Align again, so that a new component is created. Regarding the ortho-projection scaling, the ortho-projection created before update will remain in the scale in which it was created.

[Tokyo Warfare Project]

OK, I see this is at least a wanted(from you devs side ) behaviour.

I think its a limitation. At least, you should provide the options when updating what elements of the component you whant or not want to be affected so they georeferencig does or does not change.

When you come from a field run with cards full of images the first thing you will do is dump the photos, and leave the stuff aligning, wehter as draft or full to see if the run whent well.

and if all goes fine you'll whant to use already processed data by scaling, rotating and geo referencing it instead of re procesing all witch in a large dataset can cost you a workday or an overnight run.

[Tokyo Warfare Project]

Definitely scaling is bugged if you happem to do after align. I was thinking about your answer and does not match with what happens. While it seems to rotate and scale. It does not scale the mesh because the texel size remaims the same before scalling. If you set the recommended texel size on an unscaled mesh setting fixed texel sice it places absolutely massive texels. Do so on a mesh that had GCP set before alignment all goes fine and orto has neat quality. So. Please check this issue because feels to me like a major flaw.

[zuduri]

Thank you for pointing out. I will take a look at it and get back to you.