3 ms·
Looking at their prices and the (impressive) quality of the mesh topology in the demo movie they have on their web page (the rat) I couldn't help but think this
by virtualritz 4y ago
Looking at their prices and the (impressive) quality of the mesh topology in the demo movie they have on their web page (the rat) I couldn't help but think this a front that pretends to use pure AI but actually has real people (specialized mechanical turks) involved.
Specifically for guiding generation of the mesh from a possibly AI-generated point cloud (PTC). E.g. using manual contraints on an mostly automatic quad (re-)mesher ran as a post process on the triangle soup obtained from meshing the original, AI-generated PTC.
I.e.:
1. AI-generate PTC from image(s).
2. Auto-generate triangle mesh via marching cubes or whatever from PTC.
3. Quad re-mesh with mesh-guided automatic constraint discovery (think edges, corners etc.).
4. Manual edit quad-mesher constraints.
5. Quad re-mesh.
That would explain their pricing which seems a tad too high for a fully automatic solution.
$600 for 30 models x 10 iterations. I.e. each iteration would cost $2.
Or maybe it's just so niche this is simply because of number of users for now and indeed fully automatic.
Curious to hear what other people involved in 3D and cloud compute think.
- janekm 4y agoSometimes the business plan is to use human guidance initially but use it to improve the automated models, which gives you both first mover advantage and better training data… would be a smart play in this space.
- punkspider 4y agoFound this vid [1] while looking up Kaedim. The author also noticed the high prices, among other things. [1] https://www.youtube.com/watch?v=dnwDPLLzzEU