All of the tutorials that I see, having to do with perspective match, start with a stationary background and then the creation of an object to put in the scene at the same perspective. But in my case, I have an object that was photographed (a desk) where I would like to create an interior environment to put behind it at the same perspective. The environment would then be rendered, empty, and the desk would then be composited in Photoshop. Anyone know of a good way to do this, or some sort of ghost/onion skin overlay trick? Thanks. (Since I shot the desk, I would try to mimic a similar lighting style for the interior.)
Actually, seems like this article is a good starting point…
Create a plane and a front view with the reference image. Change the object properties visibility to .3 to make it transparent. Then freeze it. Any other thoughts?
My first thought was you might consider modeling that desk. It looks pretty doable. Perhaps you buy an inexpensive keyboard model from turbosquid. The rest seems like relatively straightforward geometry?
the article you linked to is based on importing a photo like you show and mapping it on to a plane to use as reference to model the object in the photo. So it is about how to use a reference photo as a basis for modeling an object versus modeling a room and for purposes of pshop in photo of object. The article describes having front, side, maybe top reference photos. It?s more difficult when your reference photo is a perspective, like you have. But still the same process is used.
fwiw I think your end result will be much better if you model the desk versus adding a photo of it over a rendering Via pshop.
Thanks for the tip! The problem is that the riser is a part that my client sells and they were unable to get the CAD files from the mfg. So I had to photograph it instead. So to model it, it would have to be exact.
So maybe this:
- Load my photograph as a background plate
- Bring in a similar object model and orientate it to match the perspective of the photograph.
- Lock camera and object
- Orient camera and object in an empty environment
- Delete object prior to final render
- Comp photograph to final render
Thoughts?
If you still have the capability to photo it you could photograph a straight on side view and a straight on front view. maybe a top view but that probably is not needed. Measure the overall HxWxD of the object. Set up a front and side reference planes with those photos mapped onto the planes and displayed at the correct, previously measured, dimensions. then model the object to match the photo references. just like described in the link you provided. you could even do it with the perspective photo you have as long as you know the dimensions of the object. You only need the overall HxWXD. CAD files would be great but often not available and this is the common procedure for modeling from reference photos. “Exact” may not be achievable but close to a reasonable tolerance like 1/2" is achievable. Pasting that photo into a rendering via photoshop will not be “exact”
Alternate approach - Model your room. Then make a plane of the correct size and map you photo to it. You will need to have a good outline around the object and an alpha channel saved in a .tif or similar file format. Make a material using the photo as an diffuse map and an opacity map. Assign and map to the plane. Rotate the plane in top view so it faces straight at the camera. Same as adding cutout photos of people, etc into renderings.
Modeling the object will give by far the best results. Much better then the alternate approach. the model can cast a shadow and maybe have a subtle reflection on the surface it sits on . Better Lighting match can be achieved etc.
Hope this helps.
This is really no different than a more typical camera match except that your image is going in to the foreground instead of the background. I have Syntheyes and it makes short work of just about any camera match, You snap some x, y and z-axis lines with at least one length and it gives you a matched camera. I quickly did one and overlayed your image onto a office scene I already had, voila. Match up some lighting and you’re good to go. I attached the camera I generated but keep in mind that I guessed at the depth of the base of the adjustable desk (18") and that the “floor” is located at the top of your white tabletop (I pulled the background scene down about 30"). It ended up with a tight 130mm lens but it looks to match up pretty well.
Cam Match.zip (94.3 KB)
Seems I’m too late to the party. I did it this way and used the white table top as the reference point.
persp_match_table_v01.zip (985 KB)
Did you use SynthEyes as well?
The missing dll should not be a problem (not sure why a .ms script would be missing though). The scene is just a camera though so you are not missing anything. Syntheyes simply creates a maxscript file that sets up the camera(s) and backplane sizes so once that’s run you don’t need either the maxscript file or syntheyes installed.
No. I used the standard 3dsMax perspective match tool.
It’s boring voice time 
the second one is not only more dialed, but also better aligned… in the first one, look at the top “line” of the monitor and the top line of the picture frames… although in the second one look how close the baseboard reveal hovers above the keyboard and how the left corner of the keyboard surface meets right at the corner of baseboard and mullion… i might move the camera down a hair so the baseboard reveal runs into the keyboard instead of hovering above it and the corner of that surface bites into the mullion, or vise versa to give a little more separation but that’s just me… great work!
Ha! I had never heard of Syntheyes until this thread and got really stoked on it (affordable too). Watched a bunch of the guys tutorials and even stumbled upon yours as well! Thanks for that!
Thanks for doing that! Seems like a really nice tool to have in the arsenal for someone with one foot still in Photography, and the other in CG.
At least I can put a voice with the name now 
Totally agree with you there. I will make the changes on the final image. I think it’s remarkable how well it works and on both Y and X axis!