Imported Camera Matching Not Working

Swizl

Active member
#1
Imported Camera Matching Not Working

So I've matched the perspective for an image backplate in Modo, but when I export the file and bring in into C3d, it isn't correct at all. I've used several different export formats that support cameras and none of them have worked. I have the full resolution image from iStock, but since I can't distribute that, I've created a lower resolution watermarked version. The files are too big to attach, so I'm uploading them to my DropBox. I'll post a reduced sized screenshot of the Modo matcher and a higher resolution one will be in the DropBoxed zip file.

DB Link

Any help would be greatly appreciated!

Thanks.
 

Attachments

Swizl

Active member
#3
Thanks Frank,

I don't know why the DB link isn't working? I tested it after posting and it downloads for me.

Here is a link to the iStock page for that image:
https://www.istockphoto.com/photo/corridor-and-lift-gm183636200-27556870
The full resolution image is 5616px x 3744px

I'm attaching the fbx zip here as it's small enough to do so.
One weird thing is that the imported camera isn't even pointing in the correct direction. I can understand some of the settings not importing correctly, but both of the world coordinates are Y+ up for Modo and C3d. So I'm not sure why the camera angle would be off?

As usual, I appreciate you taking your time to respond. :D
 

Attachments

#4
Like you I have yet to find a way to export cameras correctly, but as a workaround you can put a placeholder box at the precise camera position in Modo (including rotations) before export.
When opening the file in Cheetah you simply drag the camera on the box in the object browser while pressing the shift key which will make the cam a child of the box and also inherit its position and rotation.

The only thing left is now to tweak the FOV till it matches the background image.
 

Swizl

Active member
#5
Like you I have yet to find a way to export cameras correctly, but as a workaround you can put a placeholder box at the precise camera position in Modo (including rotations) before export.
When opening the file in Cheetah you simply drag the camera on the box in the object browser while pressing the shift key which will make the cam a child of the box and also inherit its position and rotation.

The only thing left is now to tweak the FOV till it matches the background image.
Agreed, I've never been able to get the cameras from any app to open correctly. No idea why.

Ah great tip! I'll try it out. I may be able to calculate the exact fov from the Modo info. My main issue to start was even getting the camera to sit in the correct position. Thanks! :cool:
 
#6
It's probably what Frank already mentioned: The exact details of the camera, especially FOV.

With Mis' method you should be able to match the camera position. Otherwise you can look it up in modo (including rotation).

Now Modo hasn't only Focal Length. It shows also the Angle of view which is the same as FOV in Cheetah.

If position and FOV / AOV are the same in both programs, the pic should match.
 

Swizl

Active member
#7
Thanks for the input Hasdrubal. I've tried to manually match the fields between the two, but never could get it to work out correctly. I'm sure I was doing something wrong though. I'll try these suggestions out when I get a chance.

It probably won't be until next week though when I'm back at my work computer. We sold our house this week and my home Mac is packed away until we find a place to buy and move in. I have a cheap Windows laptop, but C3d won't run on it of course.

I'll report back when I've had a chance to test.

Have a nice week-end everyone!
 
#8
I´m pretty sure that´s all not going to work because of a post processed perspective correction on the photo.
I got the floor lined up perfectly - but when I add the walls I instantly loose track - as you can see. It might be possible if the camera in Cheetah3d has a (Physical camera) perspective correction built in lens for auto vertical tilt correction.

Cheers
Frank
 

Attachments

#9
Well, obviously it doesn't work, but it should ...

According to the pdf, the photography lines up in Modo perfectly.

If there was any perspective correction, then in Modo the lens distortion should be something else than 0.0. But it's 0.

The filmback equals (a bit roughly) the 36 x 24 mm of a 35 mm SRL. It's the standard values in inches. Everything else would be a problem, but not this.

This would mean the pic was taken with a full format camera (or that it can exactly match to the format which would not be sure as with a smaller sensor the lens would have been even wider and produce more distortion. There are many professional cameras with a smaller sensor like the eos 7d (which has a crop factor of 1,6. Most photographers of istock probably have not that expensive equipment, which would mean that they have such crop factors too (with any canon, 1,6, with Nikon 1,5 if they are not full format)).

I thought, this would be standard in Cheetah, too.

The problem with Frank's render isn't that it has too much distortion, it's actually too less. The lines are too parallel for a 19 mm lens (i. e. a Zoom) which produces very much distortion, i. e. the lines are not parallel (a horizontal line is only perfectly straight in the center).

So there are two possibilities. One would be, that Frank made a mistake, which isn't very probable ... so ...

If Frank took the FOV of 84.5289, Cheetah doesn't really simulate the distortion of a 35 mm camera with a wide angle lens. It could fit with other lenses (like 50 mm and above), but probably never fully. It plain doesn't simulate an existing camera th

So my advice would be to render the thing in Modo.

If that is not possible you could fit your object as good as possible to the photo and render it without a background, then try to emulate the distortion in photoshop (it has built in filters for the other way around). Possible it is.
 

Swizl

Active member
#10
Thanks for the feedback everyone. Sorry I wasn't able to reply sooner. I wasn't able to get near a computer this week-end.

I've definitely had some problems with images that use a tilt-shift lens. A lot of the architectural images found on stock websites use them, especially when an image comes from a 3d render. Ideally I would set up a whole interior scene in 3d, but I need about 20 different backplates for an updated brochure we're working on. Modo does have a tilt-shift function for the camera, but it doesn't work well with the built in perspective matching. There is a work around to oversize the image canvas, perspective match, shift the camera and then redo the perspective match. Not the most elegant solution, but it does work. There is also a function to adjust the camera distortion, but most of the plates I've been working with haven't been bad enough to need that adjusted.

I've been back and forth about just doing all the plates in Modo, but the shader tree is a lot more difficult to organize IMO than the C3d node based materials. I haven't been able to figure out how to replace materials on imported objects and also how to just delete materials attached to objects without deleting them from the whole scene.

Thanks again for the input. I'll just put a lot more trial and effort in testing different methods. I think if I get close, that will be good enough. It's pretty much how I've already been doing compositing in C3d, is guessing at the FoV matching anyway. I'm certainly more comfortable working in C3d. I feel like I'm fighting Modo most of the time. The Action Centers and Snapping are confusing at best.
 
#11
This is just not Cheetah's cup of tea. The Shift-Effect is something else you can't really simulate here, like film-back and the lens correction.

Like I said, you can get (or get rid of) any distortion in photoshop you like. But it's tedious, and regarding this pic you don't have any other thing to do then render in Modo.

Modo's shader tree is something different, and you're right, it's not that easy to grasp. Modo has shader nodes, too, but they seem to be a bit difficult, while the basic layer system is actually even less complicated than Cheetah's nodes. But much more powerful as you can stack up the materials as much as you want (well, it can get a bit more difficult with masks, texture variants and stuff). You just have to understand, it's always from bottom to top.

And too change a material: Select the polys (or alternatively nothing for the whole object) and hit M (and give a name or choose an existing material). If there, an old material doesn't get deleted.

That you can delete in the shader tree without affecting the model.

It's really a quiet simple system, but very powerful (and it doesn't seem as if the node based shader system ever really took of. The layers are enough).

If you want a material with another shader (like no reflection rays or so) you put the material above the base shader and a new shader (without the reflection rays or whatever) atop of this.

Modo has quiet a big material library. You could use pbr, too, so even if it is at this point too much to learn to create your own materials, you have a wealth of materials to choose from (pbr would be good, if you wanted to render the object in Cheetah, too ...).

All in all it seems a lot of work to match an existing photography in any 3d app; it's maybe more easy in post with after effects or so (about that I don't know plain nothing), but I think you'd have to do it a few times to really master it.

And then comes the light which often is a giveaway ... wrong color, wrong intensity, sometimes even wrong direction. Next time I'd do it fully in 3d (as I said before, once). Actually you would have had far less work for this 20 pictures than with all the camera matching problems you seem to have had.

All you would have to model would be the rooms, maybe a window or two, and (maybe) the lights. Everything else, like chairs and desks, you can get for free on different sites (as you have modo, it has a model library, too).

Even if you say, you don't know how to do it, in all the time you used for this matching stuff, you'd have learned to model this things. And well, something always worth considering is: Which way to get a result is more fun :smile:
 
#12
Hi Swizl

Just remembered. Not so long ago I stumbled about a few tutorials on modo from richard yot, only a few of them free because they were made for much older versions: Exactly those that would help you with the materials, especially that one about rendering interiors and rendering products (at least the latter seems to explain the shader tree from ground up; had only just a look into it, though).

https://gumroad.com/richardyot

I hope, this will help you with your current problem, even if I'm a bit late with that.
 

Swizl

Active member
#13
Hi Swizl

Just remembered. Not so long ago I stumbled about a few tutorials on modo from richard yot, only a few of them free because they were made for much older versions: Exactly those that would help you with the materials, especially that one about rendering interiors and rendering products (at least the latter seems to explain the shader tree from ground up; had only just a look into it, though).

https://gumroad.com/richardyot

I hope, this will help you with your current problem, even if I'm a bit late with that.
Thanks! Yes, Richard Yot is a rather brilliant guy. I've watched some of his free videos. He's highly regarded on the Modo forums too. I may buy some of his paid for tutorials, since they are suppose to be really good.

There are a few other free tutorials on the shader tree that I've either watched or have set in line to watch. The shader tree is really powerful, but it can also be extremely confusing. Especially when something goes wrong.

I've come to a point where I was able to sort some of these issues out with some work arounds. Making render passes in Modo is pretty convoluted. At least compared to C3d. Rendering out a shadow pass isn't as easy as it is in C3d where adding a render tag and setting the background to transparent then hitting render is most of what you have to do.

One of the problems I've come across in Modo is enabling Environmental Importance Sampling makes the renders look a lot better, but then it also renders all of the scaffolding geometry that I only intended to catch shadows. I think setting EIS to on is basically turning on unbiased rendering. So no way around it but to render the shadows separately and then composite in Photoshop. A few of the more straight on shots I just cheated anyway and used a drop shadow effect in PS.

I was talking to some of the wiser people on the Modo forums and they highly suggested taking my own shots for compositing. It's way too hard to control what you get from a stock website. The images are sometimes cropped, and typically have the exif data expunged. Then there is an issue of matching the lighting and color after that. A large majority of the interior renders/photos also have a tilt-shift applied to them to straighten the Y-Axis. They recommended using one of these 360 cameras to make my own hdri that will match a backplate. I'll start with the lower end model and if it works out well, jump up to the HD version.

Ricoh Theta 360 HDR Camera:
https://theta360.com/en/about/theta/s.html
https://theta360.com/en/about/theta/v.html

It will probably be what I do in the future, but no time to do it now that I've already started on the backplates. I've gotten about two-thirds of them done now. I've been using a very neutral HDRI in Modo and then applying a color processing effect on the hdri to color cast it similar to the backplate. As much of a pain as it's been, this process has been a good learning experience.
 
Last edited:
#14
Great to read, you have all problems solved and at least are able to do what you need (still a lot of work). I see, my answer wasn't needed anymore.

Didn't know that you can take a sphere-HDRI as backplate.

I'll start with the lower end model and if it works out well, jump up to the HD version.
Actually didn't know that cameras like this are already this inexpensive in good quality. As lenses usually are very expensive (or not that good), it'd be really interesting to see some of your self-made HDRIs.

If I'd plan a trip at the moment, I'd be very, very tempted ...

And thank you for your in-depth-answer. Very interesting.
 
#15
But before you buy, I have one question for you.

You do need the end result in 4k or even bigger?

If so, the resolution of this camera probably isn't high enough to get a sharp background. From the end picture it creates, you render just a part, depending on the 3d-camera angle. Say, you need just a quart (could be even less), that's with 4 k not nearly enough, nor the still image resolution of 5376 x 2688 px.

Some 1100 px would be stretched to 4 k. Even if you render in full hd, it wouldn't be enough.

Just test it with a hdri you have with approximately the same resolution. If you can get with that the quality you need.

Or I'm completely wrong here? Otherwise, sorry to be the spoilsport here.
 
#16
..... It's way too hard to control what you get from a stock website. The images are sometimes cropped, and typically have the exif data expunged. ....... A large majority of the interior renders/photos also have a tilt-shift applied to them to straighten the Y-Axis.
I´m pretty sure that´s all not going to work because of a post processed perspective correction on the photo.
and cropping of course.

Surprise. :p ;) :D

Cheers
Frank
 

Swizl

Active member
#17
and cropping of course.

Surprise. :p ;) :D

Cheers
Frank
Oh yes, your advice was duly noted and still as always appreciated!

I did cheat some though since I didn't need all the shadow catching geometry to line up perfectly with the walls, I just let it stay uneven. In one instance I forced the geometry to fit, even though it technically wasn't "correct". I guess the 3d police will have to come get me and stick me in a linear subdivided cube! ;) :cool:

I guess my ignorance was bliss before I found out how I had to go down the "wrong path". Don't tell my mother. LOL
 

Swizl

Active member
#18
But before you buy, I have one question for you.

You do need the end result in 4k or even bigger?

If so, the resolution of this camera probably isn't high enough to get a sharp background. From the end picture it creates, you render just a part, depending on the 3d-camera angle. Say, you need just a quart (could be even less), that's with 4 k not nearly enough, nor the still image resolution of 5376 x 2688 px.

Some 1100 px would be stretched to 4 k. Even if you render in full hd, it wouldn't be enough.

Just test it with a hdri you have with approximately the same resolution. If you can get with that the quality you need.

Or I'm completely wrong here? Otherwise, sorry to be the spoilsport here.
Yes and no. We do a pdf version (lower resolution as they are meant for on screen viewing) and we also do a print version of the same catalog. Most of the shots though only need to be around 300 dpi at approximately between 3" – 6". Some of them may be bigger, but for the most part we are fitting several shots on the same page.

Here's the original one I helped work on 5 years ago that's going to be updated if you want to get an idea.
http://www.arrissigns.com/documents/catalog2015/ARRIS-2015.pdf

So the 4k camera most likely would fit our needs for the most part. We have a few high end digital cameras that I could practice creating some HDR's the more manual way and then stitching them together. It will all be some trial and error and practice to get to a good workflow. But that's part of the fun. It's more of an issue of time right now.
 
#19
Thanks, Swizl, for the further information. I'm looking forward to see some of your tests. Both with the possible new cam and the stitched HDRIs (some work I'v always shunned).
 
Top