Tweet
hi i am having great results with zephyr when scanning objects!Now i am trying to learn head scanning with masquerade.My results are acceptable but not very sharp, i am using Canon Rebel XT and was wondering which settings i should change in zephyr advanced mode?thank you
-
-
Hello,
depending on the images it might help to change the noise filter redcution and set it to use raw results rather than refined, but this may vary. We are actually going to insert a body preset in v1.500. Also consider that close range works better when the pictures are taken roughly from the same distance (i.e. the three classic orbits).
If you'd like to share with us your pictures i'll gladly have a look at them and i can also send you the presets we're working on.
Andrea -
Hi ,
I'm trying to scan human ear using Canon S120 or Sony DV camera with mpg output.
Firstly - Zephyr does not recognize mpg video from camera. It does not read it at all.
Second. I've not got success processing photos of ear. Zephyr displays message, that something gone wrong. If would like to obtain some sensible results before I buy software.
Photos are here.
https://drive.google.com/file/d/0B1SfuzAzjy8dV2NoT29rS2ZwNGM/view?usp=sharing
Setting I used were "Human Body - Default" or "Close range - Fast" and combinations of these.
Please comment what to do.
Comment
-
Hello pacak,
thank you for your feedback! I've seen the dataset and probably the lack of context (the cardboard cutout) is actually throwing off zephyr rather than helping it during the SfM phase.
I will send you privately via email an ear scanning sample, so you can try and replicate the shooting method we used.
Zephyr works by understanding features in the image and then matching those features in between other images. By using that cardboard box you are actually removing features from the images since there are not easily recognizable patterns. An option could be to actually use the masking in 3DF Masquerade for the second phase (but not in the first) or (only if you don't need the color, but just the mesh) to use a pico-projector to project a pattern directly on the ear.
Your dataset would probably work at least using deep/exhaustive settings (but will take a long time to crunch the numbers) or a custom preset.
Also, could you please send us a video sample? That way we can replicate the issue and work on a fix
AndreaComment
-
Sorry for the barrage of questions.. I am just looking at different software available and comparing..
1) Does Zephyr use the GPU for processing the photographs and generating a point cloud? Or just for displaying them?
2) Can it use the GPU and CPU at the same time?
3) What is the best CPU you would recommend?
4) What makes your software stand out above the others?
Many thanks!Comment
-
HI there
Any idea where I can find 3DF Lapyx, the camera calibration routine?
Link says page not found.
Thanks
Hennie -
HI Andrea please mail me at [redacted] so that I can send you the link to my dataset.
Tks
Hennie -
Hi Hennie, just sent you an email and edited your post to keep your privacy safe from bots!
-
-
Hello Quantumsheep! No problem, feel free to shoot!
1. Zephyr uses GPU(s) available on the system in all the phases of the computation. Structure from motion phase (point cloud generation), Multi View Stereo (dense point cloud generation) are all GPU(s) accelerated. There is just a small sub-phase at the end of the SfM phase (clustering) which, by nature, can not be run in parallel, but that aside, zephyr pro uses as many video cards as it can in order to spread the computation cost (zephyr lite is capped to 1 available GPU). Please note that only nvidia cards are supported.
2. Yes, when using GPUs the CPU is used as well!
3. I personally would suggest from an i7 upwards, but a rig with an i3 (for example) and a high number of video cards would work better than an i7 with no cards, obviously =) Remember also that an SSD and a high amount of system memory is highly reccommended.
4. Above all, I would say precision and reliability (though this sometimes comes to a performance cost, and not all case scenario require high accuracy. We do scale lineraly for the MvS phase though, which can't be said for some of our competitors, which are slightly faster for smaller dataset but for bigger one (300+ pics) are absolutely slower). Our technology is completely built in house and is the result of years of R&D, which means that we have (and give the final user) the complete control over our reconstruction engine (pro version) allowing the user to touch a REALLY high number of parameters (usually presets work for most cases, but then again, advanced users love the possibility to fine tune every aspect). A big plus is also our masking tool (3DF Masquerade) that comes bundled for free with both pro and lite versions of zephyr (make sure to check it out here, but it's much better if you try it out yourself, it's also included in Zephyr Evaluation ) . Zephyr is also made with at least 30% more love. Enough with the commercial break, let me know if there is a specific aspect you'd like to shed some light on!
Feel free to ask more!Comment
-
What a fantastically detailed, speedy response!
Thank you very much I will have to check out the evaluation copy when I have some free time.
How about an i7 rig, 32gb ram and 4 GeForce GTX 970 cards?
I have noticed that some of your competitors even suggest that 32gb isn't enough to run their software with high res pics!
So what camera would you suggest for your software? Any particular lens seem to work better than others in your experience?Comment
-
That sounds like an awesome rig =)
In my opinion, 4 high-end cards are worth their money if you are going to need very fast processing. May I ask what kind of dataset are you planning to run ?
I personally have my home computer with 16GB of ram that works very well even with fairly big datasets. Our test workstation use quadros and have 32GBs of ram and from the test we've run our memory footprint is generally lower than our competitors. Also, as far as we know, some of our competitors load all in ram to be faster (and it is true, it is faster), but the downside is that they hit their memory limit very soon, while we do dump a lot of temporary data on the disk.
Well, obviously DSLR cameras are reccomended =) in my experience we're usually more accurate with close range objects. We're agnostic to brands or lenses though, as we just use raw image data, so the better the camera/lense, the better the result! Remember that rather than a camera with a high megapixels count, you have to check the sensor size and the lense quality.Comment
-
We recently started capturing the beautiful statues around the mountain dwelling temples in South Korea. We thought it would be cool if we could actually capture the temples too! So need to do some serious research into the technology, as so far we have just been using free versions and 123d catch ~ which I have grown to really not like very much! haha ^___^
We are planning on doing Gyeongbokgung Palace, in its entirety, in Seoul.
We have a Cannon g15 that seems to work well. We are also looking at filming in 4k, because I saw you have a feature that allows you to extract frames at intervals from the footage.
I am thinking about making a servo driven attachment for the camera too.. This could be cool for high shots (Might make an extra tall, telescopic monopod with this servo driven mount on top) So many ideas!Comment
-
One other question. Will there be any future plans to export normal maps, bump maps, etc?
It would be amazing if you could generate the diffuse map with different material id's, and then export the UV coordinates as several maps, rather than one big one too. Something all the software I have tried lacks. (I am only presuming, as have not tried yours yet!)Comment
-
That sounds really interesting =)
A canon G15 sounds good as a starting point. Honestly i don't think 4 GTX 970 are worth their money if you're running 70-80 pictures datasets (which i'm assuming for statues) but might come in handy if you're going to process whole temples (depending on the accuracy you want to achieve).
As a simple comparison, i just ran piazza Bra (main square in verona) with 330 pictures (7Mpixels) in little less than 10 hours at default parameters, with an i7 / K2000 / 32GB.
In your case, I would probably use a DLSR for aerial photos though.
As for shooting in 4K, i have to admit that at the moment there aren't specific optimization when extracting frames, so be careful when extracting frames from huge videos (it's on the todo list =) )
Normal maps might be added in the not-very-near future as it has been on our plans for a while but it isn't a priority though. Same as for UV coordinates as several maps, we might do that in the future, but since this is usually a feature required by professional 3D modelers, since they usually are much more comfortable of doing the post processing in their editor of choice, this isn't a priority either.Comment
Comment