Processing photos is at least as important as taking good photos. Unfortunately, finding the right photogrammetry software (especially for Mac) can be quite an issue. And even with the right software (and of course hardware), it can still be quite tedious to learn the usage of such software. For this reason, I started working on the OpenScan Cloud over a year ago. I had quite a few setbacks but over the last few weeks I had some amazing progress, that I want to share.
Raw (!!!) scan directly into printing - it took 2 x 3 mins to take the photos with the OpenScan Mini, 10 mins to finish the reconstruction (automatically), 1 min to scale and slice and 4h to print ...
First tries in 2020 (and a little rant...)
After talking to several people, I teamed up with some skilled programmers that seemed to follow the same vision. We quickly settled on the major layout and started to distribute and work on the different tasks. As you might now, I am not very experienced in programming, and thus I had to rely on their guidance. Over the course of the next few months, we made quite some progress. The processing was mostly done in Meshroom and Meshlab, both great open-source tools. For this reason, I was quite happy that there would be a publishable version by mid 2021 or so. and next to those already existing opensource tools, we created some quite useful new tools, that should have been published already…
One thing that I have been especially proud of is the auto-scaling feature for the OpenScan Mini. Basically, you take all the calculated camera positions that all lie on a sphere, calculate the radius by using some math-voodoo and scale the whole model accordingly. I did a lot of testing and found out that this method can yield great accuracy (somewhere in the range of 0.05-0.1 mm on a 5 cm object). As far as I know, this would be one of the very few (or even the only) and very accurate marker-less scaling methods for photogrammetry. In combination with the cloud processing, this would bring the low-cost OpenScan devices on par with those 3-10k+ structured light scanners out there...
BUT the whole progress came to an end, when my "partners" decided to go fully commercial and especially fully closed source and of course all written codes would be their intellectual property. (There might be a chance of fighting that, but I neither have the mindset nor the money to go into such a stupid fight)...
So, I had to restart, because I still believe in this idea and want to make it happen. And so I moved on…
One year later - Apple Object Capture API
Raw and untouched scan from Apple's Object Capture API
In the meantime, Apple decided to publish an open and free photogrammetry API (but of course not open-source). And to be clear, I am no big fan of Apple and have so far tried to avoid their products and services, BUT after seeing several reconstructed sets, I had to give it a try.
I have been playing around with Object Capture for a few weeks and wow, it is just great. The results are crisp, and the reconstruction is very robust, accepting even some very low-quality photo sets that no other program was able to reconstruct. And of course it is very fast ...
Downside (as usual with Apple), there are not many options to customize the reconstruction and there are almost no parameters to tweak and the output is only the 3d model (which is great), but no additional data like camera positions or any metadata...
Object Capture - Stress test
I have re-run over 10.000 photos from almost 100 different scan objects to stress-test the software. And the results are super promising. I was able to reconstruct 95% of the image sets, whereas other programs were able to reconstruct only 50-70% !
The reconstruction times were significantly lower, and the overall output quality is great. Unfortunately, due to the missing metadata, I won't be able to do the magic auto-scaling...
Anyway, I still have to do some more testing and as the API is still in beta, there might be some changes in the near future.
OpenScan Cloud - What will it be like?!
I see people using the OpenScan devices in several ways. Some are going the full DIY route and prefer local processing. And there are great commercial programs available, and Meshroom is a good open-source alternative.
But there are quite a few people out there, who would prefer an easy-to-use one-click-solution. And this is where OpenScan Cloud comes into play.
I am currently setting up a scalable and hopefully robust solution, where I intend to make it publicly accessible. With my current setup and some basic usage calculations, I should be able to reconstruct ~ 5-10k photo sets per month. It will be token-based system, where you will be able to use either the public token or an individual token to do a reconstruction. (Individual token will allow some more personalized features). The feature will be directly implemented into the OpenScan firmware, so that you can either process the set locally on your computer or upload it and get the result by email and/or direct download link.
The backend is already 90% done, and I am working on it almost every day :)
I hope that the documentation for the access point will be ready by next week. Maybe I can find someone trustworthy who could develop a simple GUI, that would allow people to select and upload files directly from their computer. This should not be too difficult, but I will be busy with implementing this feature to the OpenScan firmware and thus need some additional helping hands.
So feel free to reach out :)