Turning Mobile Phones into 3D Scanners
Demonstration from the Computer Vision and Geometry Lab (part of ETH Zurich) of an unreleased app which can capture 3D data with a mobile phone camera without the aid of a depth sensor. Video embedded below:
There is no documentation currently available online, but it appears that data generation occurs both in-phone and with the assistance of data-connection, as well as being exportable.
TIL there’s an online (underground) marketplace for selling animal crossing villagers…
Reversal works correctly for sentences of length > 1.
In the example above are three separate sound files, where I say “Hello”, “How are you?,” and “Goodbye,” respectively. Changing and reversing playback frequency preserves the apparent position read from the stream. For example, if I say “Hello”, reversal should say “olleH” and then go to “eybdooG”, etc.
EDIT: Large mp3s now work as well
Replicated FMOD’s reversal function. Gapless; sounds the same as if FMOD were performing the reversal.
Fine-tuning is necessary to deal with formats other than .WAV, though. But it’s a start!
So after a long time fidgeting it seems FMOD has a bug with reversal of sound sentences, where it spits out garbage for every subsound other than the first.
One possibility to fix this and ‘cheat’ reversal is to preload the reversed version of the sounds loaded. Since the gapless sentence + variable playback frequency features still work going forward, I think we can load a second channel with the reversed sound samples, switch to it when rate drops below 0, and just go ‘forward’ on it by flipping signs. There’s some position-setting that needs to be done beforehand, though, when the flip happens. I’m holding out hope that that FMOD feature will work correctly.
Above is an example of my algorithm reversing samples.