POINTCLOUDS from 2D images (+depth) - 2020-02-03 11:02

POINTCLOUDS from 2D images (+depth)

Link to main site

There is some problem with zip:
“le dossier zip n’est pas valide”
I have tried 3 times

This is strange… Some other people successed opening the zip.
Let me check. Are you on MacOS?

.zip fails for me as well, on Windows using the built-in Windows Extract.

I was able to successfully extract and access the files using 7zip, but something about that .zip is ‘non-standard’…

Ben, yes the first one is a .rar just renamed as a .zip because the asset uploader don’t take .rar extension.
The second one (on my dropbox) is a fully standard zip.
In my computer, both versions are OK to open and extract.
Is it better with the second one? (I will replace the asset if so)

great work!
Just a question:
how do you predict depth from 2D images?
The only way I can do this is with DenseDepth, is there any simpler method?

1 Like

I think it might be better to use the .zip so others don’t have this issue. Thanks!

1 Like

.Zip did work for me, although it complained about actually being a .rar or something like that. Anyway, this is amazing. If you want an easy 60 fps realtime, just change the point type in the line MAT to circle (sprite).

1 Like

Thanks. In this particular case, I manually & fastly paint the depth in Photoshop (3 or 4 minutes each, really). For this effect, it can be very rough and approximately done.
And of course you can use color + zdepth from any 3d sofware.
It propably exists, but I’m not aware of any method to get proper zdepth from just color.

Hello, yes it was a naming problem, OK on mac, ok with 7-zip, not ok with windows standard unarchiving…

1 Like

ok, I suspected it was a manual approach.
At the moment I’m using DenseDepth Machine Learning algorithm to extract depth maps from images,


something like this

370322_140413_124056109-ad44988a-fb54-4a2f-bfa1-c09b1154791c DenseDepth - February 3rd 2020 at 6.57.05 PM

I have a 3D point cloud.

Now I want to convert it to 2D depth image of say 240*320. Does anyone has anything to share?

It will be a good addition to Matlab.