garageCube

I am trying to calibrate an OBJ of a very large structure (a large bridge in a city centre). Right now I am just trying to setup a rough perspective under the bridge as a sort of "pre-vis". The issue is MM seems to load 3D objects with a totally flat, isometric perspective and you can not manually position a camera to look - for example - inside a room or under an object with perspective. I rendered a perspective POV from the 3D model and tried calibrating points based on that. Once I have all 6 points lined up very accurately, the model just disappears. I should be looking under and down the length of the bridge model with the object going overhead and behind me.

Whats going on here? is this some sort of camera clipping issue if a model extends out past and behind the camera in MM?

A basic camera placement tool in MM would be really helpful to use 3D objects...

inside a room is not ideal situation.
try with an object

meanwhile you can post your scene here for inspection

Here is an example of the type of perspective I am trying to calibrate - a view from under the object (the bridge).


Here is how things load into MM:


Note, I am just using the background image as a reference to try position and align the 3D model.

Each time I try and calibrate, the model just disappears. If I play with the calibration handles and try editing it, I can get it to flicker in and out but the perspective is not anything close to what I am trying to line this up to - even though the vertices are lined up very cleanly on the reference image.

thoughts?

post your project files (background + 3d model) and I'll give it a shot.

Thanks Franz, any way I can share the files privately? I'd love to share the outcome of this with the community but I am not permitted to share the 3D model publicly.

either send your original files to the support at my attention, or post here a simplified version that you are permitted to share .

Will do! Thanks.

OK, I've had a look at your files.

- 1 make your render the same size as the output: 1920*1080 for instance. make sure the ratio is the same
- 2 crop your 3d object to the visible parts. your bridge is very long, to calibrate on your view you don't need half of the bridge
- 3 make a rendering without the people, this will help you to set the points at the right place
- 4 make more remarkable points : maybe subdivide the top of the bridge
- 5 make sure your calibration points are NOT coplanar: they should be all on a different plane.
in your mesh, I have problems finding non coplanar points
- 6 while calibrating, you can still rotate your mesh by pressing the "r" key, this will help you select good points
- 7 as a sidenote, your mesh seems to be different from your 3d rendering: the windows do not exist on the 3d object,and the back wall doesn't have the same proportions.

let me know how it goes, or send me(through the support) updated files

On the other hand, I don't think you need 3d calibration to map that bridge .... it will be easier to map it manually.
Usually, 3d calibration is for contained complex objects, not minimalistic large scale building.

Awesome, thanks.

You are right, I don't really need to use a 3d object to map. All I am trying to do is set a view in mad mapper for the purpose of a simple pre-visualization that approximately matches the POV and FOV in the image. Currently the only way (that I know of) to change the FOV on a 3D object in MM is to calibrate it. It would be super handy to have some basic 3D positing tools in MM - a basic camera placement tool, etc.

I'll try your recommended steps.

FBX loading with camera support "should" come in a future version of MadMapper.
We're evaluating the proportion of our users who actually know anything about 3d.

On the other hand, to previz your mapping, just load your rendering as a background , add some quads/surfaces and you're good to go.

Cool, that would be great. I totally understand that not everyone is using a 3D heavy workflow... its what I love about MM!

Previs wise, I'm really trying to leverage the ability to map a syphon feed to a series of predefined UV maps as a way of pre-vising video textures on the bridge. thanks again.

I have the same problem as this user but unfortunately I can't use an alternative method. When trying to 3D map the model disappears or there is an incorrect angle and FOV.

Given the software is sold on it's ability to 3D map, shouldn't there be some better documentation on this?

Hey, FWIW, I ended up just doing all of this (previs) in Blender. The 3D functions in MM are cool but really aimed at a very simple setup: 1 projector, fairly small model, medium to long projector throw. My suggestion is that if you want to run a complex project in MM that is built from 3D assets that you build it in such a way to deliver as flattened 2D assets (IE baked texture maps, etc). MM is fantastic for doing complex things with 2D assets that you then map to 3D objects. The native 3D functions of just too crude for most large projects, however.

best,

A

Thank you - I think I've learnt the hard way this is the case!

Really enjoying MadMapper - very impressed with the DMX and edge features - but just doesn't seem able to do what I need it to do.

I'm using a complicated mesh topography .obj with a simple orthographic top down UV (ie. the default) - tried to align peaks and troughs but the error margin is still too great to calibrate.

Alternative I'm looking at is OpenFrameworks or TouchDesigner - or just positioning the projector perfectly centrally at a 90 degree angle to the physical 3D object that I can use a plain mesh...

I am assuming that there is no way to make sure a 3D model always works? I've been running into this issue quite often.

Are there any steps that can be taken to make sure a model has the highest chances of being compatible?

I saw the comment about keeping points on different planes, this seems to help, however, not all models have over 6 different planes of depth, some might only have 2 (a geometric repeating pattern for instance).

This issue seems to be consistent across platforms for us. We go between MadMapper and X-Agora and this same issue plagues both, which makes me feel that the issue is within the models perhaps.

Thanks!