AI Makes 3D Models From Photos | Two Minute Papers #122

37 thoughts on “AI Makes 3D Models From Photos | Two Minute Papers #122”

1. ommggggggffoddddddd !!!!!

2. I thought I understood what this was about but then he started adding chairs together….

3. How's is this not an application of structure from motion, or one of the many other iterations of machine vision?

4. Im early

5. 2:10 "i am so happy to see breakthroughs like this in machine learning research". Why? THERE WILL BE NO JOBS FOR 3D MODELERS!!!!!!

6. I wonder if by the end of the century there is still anything left that can't be done with machine learning. It's basically everywhere you look already. I can't wait to see what's next 😀

7. Next full scene reconstruction of scenes. I start to understand those who fear technology

8. when this is refined it will be amazing…

9. this is awesome!
   You know what I'd like to see though? Trying this basic idea on light field recordings.

10. Does anyone here know how you could construct a 3D model of an object using video, if you knew the exact location of the camera in 3D space (but not the location or distance of the object)?

    I need some starting points to begin my searches and some good papers to read.

    Thank you in advance!

11. I wonder how different or how much harder a vertex-based approach would be, to create more perfect shapes

12. But can it tell the difference between a donut and a coffee mug?

13. Holy SHIT somebody just made Punchcard alchemy from Homestuck an actual thing

14. wonder if this app can be used as a pseudo-depth sensor for monocular systems like standalone security cameras or small cheap drones. allows it to build enough of an internal model to navigate its surroundings.

    also feels like this can be the first step in a match moving system, like those raw "scans" are meant to be a sketch or template for the app to build off of rather than being the end product. like it's the raw volumetric data the app gleans off the photo that it then compares and contrasts with more refined meshes in its database.
    there is an app that debuted around 2013 called 3-sweep that let you trace 3d models from 2d images by way of an automatic edge detect + symmetrize function and a year after that a similar app that let you overlay and conform existing 3d models over 2d images. bet this new app shown could be assisted by the former so it could perform the latter if I make sense.

    amazing stuff regardless.

15. This is very human

16. I am so glad, that I found this YouTube channel 😃 it is incredibly interesting to watch! By the way, what job deals with those topics covered by this channel? 🤔

17. Watching these videos make me so happy! What a truly awesome time to be alive…

18. As if we didn't have enough problems with copyright with 3d printers.

19. I'm guessing that there is no point into going into a career in 3d modeling and animation now.

    I wonder how much longer it will take until 80% of humans are obsolete and unemployable.

20. Oh the possibilities!

21. Best channel I have found so far in this year..it's so sad that those kinds of chanels don't exist in german so I could show certain videos to friends and stuff without translating which is pretty hard at this level of complex themes. Keep up the good work

22. the next step is having it develop a 3d model from pictures of the same object from multiple different angles/lighting conditions, and then extracting a full model from a video!

23. Man oh man! All of these machine learning papers recently….. It's giving me the idea, realistic or not, that an artificial general intelligence is closer than I thought- By a good margin perhaps….

24. 1:34 The car-boat melding forms a submarine shape.

25. I love this channel

26. Well, it seems like a step up from photogrammetry. Very very interesting! Loving this channel

27. After watching many of your videos, I think it is reasonable to expect that by the end of the year as similar algorithm will be able to create precise, high resolution 3d models. It blows my mind.

28. Huh, getting more out of less!!!

29. OK if this is possible from a single image, why aren't we getting photogrammetry Deep Learning software that does it in a tenth of what the rest do it, with massively less noise and much more precision? It seems like the perfect match for DL ML.

    Point clouds could be done much faster and the results better. I just want to cryofreeze myself and wake me up pls when this is available

30. I will see you next tiiiiiiimmee

31. Generally ok when it comes to models for decoration purposes but when it comes to animating, skinning, bending the model or rendering it with a realistic render with raytracing, the outcome would be horrific. Also there's no requirement for scanning simple objects such as cars and chairs which can be easily modeled with polygonal tools but regarding rocks, stones and stuff like that , they can be scanned and then retopologized with a new mesh.

32. My comment is simple:
    Holy fucking shit.

33. Here from the latest video, we've come a long way, doctor!

34. Thank you for making useful research that may not be so easy to find, making it quickly understandable. I just recently found this channel, it seems great. C

35. "Absolute Madness" lol

36. I would like to see higher resolution models with stereo cameras and bino vision to aid machines with spatial awareness while keeping low costs

37. Any more current models now?

Comments are closed.