How to export tracking data in some format? So it later could be imported in some other 3D editing tools

[ArtemBernatskyy]

Thx for great software!

How the mesh could be exported in some 3D format? So it can be later imported in Blender, After effects or similar?

[aharley]

Interesting idea. To clarify, it probably won't be a "mesh" in the sense of a 3D object (since our output is 2D), but something good should be possible here.

I'm not an expert in Blender/AfterEffects or similar software, so maybe you can tell me: what's a good format? Can you point to an example of something similar?

[ArtemBernatskyy]

"since our output is 2D"—huh?
I thought that somewhere in an intermediate step it should be 3D...

But if you're saying that basically it's 2D the whole time (i.e., the neural model outputs 2D coordinates), then it's not possible to convert to 3D without modifying the core technologies, which is a hassle.

speaking about formats it's probably any of:
USD
Alembic
FBX
glTF 2.0 / GLB
BVH
C3D

[aharley]

Yes, 2D the whole time.

[bhack]

@ArtemBernatskyy On another thread we are discussing to extend the GT to 3d coords for the mixed datasets assembly used by All trackers (where it is available or it is still possible to compute it). See #8 (comment)

If @aharley is going to update these on HF it doesn't mean directly you will have a 3d output from the network but at least with a 3d GT you can eventually fine-tuning/experimenting your own ideas for a 3d output.

Extra annotations dictionary keys in GT like camera Intrinsics/extrinsics could help a lot if you need to learn it as at least you could need to inference and export a camera with its parameters if you don't use a canonical one for this party apps.