Cannot Project a cam for all of my photos shows up unaligned, scan is TERRIBLE

[Jinny]

Hi,
I'm in a dilemma here trying to figure out why it isnt working, see link.



I have a camera shot of a head with 5 views all around model 180 degrees, complete with EXIF data, it was taken with a digicam. The resulting generated mesh is far worse than my previous scan which was only 3 photos of a girls head from 3d.sk with no exif data despite using the same settings. see sample here,



I even did a second test on the guy's head but this time with 11 photos, i get the same bad result where photoscan only projects 3 out of 11 photos with a camera, the rest are not aligned with no cam and is totally purposeless. see link here,



I also dont know why the planes(cams) generated from each photo does not position around the point cloud properly, and
sometimes it is even facing the wrong way, or facing as a row. often some of the photos
were not even generated a cam and show up under the chunk as NA(not aligned).
 
I am guessing the only way to fix this is to calibrate those photos manually, but i have no idea which values to put in under camera calibration so that when i align my photos the cameras are properly positioned like they should be. Anyone can direct me instructions on how to
input calibration values correctly, just wished there was a grid in the scene telling me which point in the 3d space for which xyz value, or better yet a way i can simply move the projection planes around in the scene into the correct place?   

Apart from this i often get cannot build geometry error on some of the photos captured even if it were from the same camera.

If you want i can even send you the pics over at dropbox and my scene as well so you can download and inspect the EXIF of the photos. Just PM me or leave a message here.

I am using Photoscan pro edition, ver .9.0 build 1542 64 bit running on windows 7.

[Wishgranter]

First of all, photos what you show here have NOT same resolution !!!! 2nd, pscan need that photos are taken with multiple cameras and not with one moving around object, because the model is moving even slightly. For better understanding if you use 18 Mpix camera and shoot from 1m distance then you go with precision of aprox 20-50 microns so thing about it, its mostly unpossible to shoot someone with one camera an object will not move - 50 microns are something about hair diameter !!!!

So using 3D.sk reference is unpossible, they shoot with ONE camera so object has for 1000 % in move between shoots...... best if try shoot personaly, with explaining the person not to breath for few seconds, be totally quiet....... OK ? 

[Jinny]

thanks for your reply, even if you seem reprimanding. I am in no means boasting myself.   

Pardon the resolution, its photobucket rezing my oversized photos automatically to lower resolutions, but regardless you can see everything clearly and that is good enough .

as for 3dsk, id be happy if i could get at least the same result i did with girl face. I'm a modeler and i just need photoscan to produce a good enough base for me. The thing is that girl face is like the only one from my 3d sk collection that seem to have generated pretty well, all others did not which is why i am left wondering??

I see your point in having multiple cameras, but there is no way i'd be buying 11 cameras or more just to scan a head. would it make a difference if i orbited around the model taking shots of him from all angles instead of him simply rotating while i leave the camera steady? IF you see the tutorial pdf in their site, they have a statue in the outdoors, i doubt that one was taken with multiple camera rig all around the model instead of just one, but yet their scan was great.

[tincansassoc]

Assuming your model can stay very still, try walking around your model whle taking the photos every 15 degrees, and do this again looking up and again looking down. Im guessing 30 to 50 photosmight do it... If that doesnt work, then use the masking tool and completely crop out the face in every image, this will help camera alignment a little because of the subject moving relative to the room... photoscan references everything in your photo, so removing the room from calculations will help alignments. Its just a lot more work and dont expect perfect results. Your current photos might work a little better if you masked out the background, however you should be using more photos too since photoscan needs to have overlapping images. That is also described in the photoscan documentation...

[Jinny]

thanks for the tip. I did try masking the 5 photos, still same result. I actually began using 16 photos with overlap in total, when i first scanned that one out, it gave me a poor side view of the face and only had like 10 out of the 16 actually show up as camera projections. the rest were unaligned and pretty much useless. Then i went for 11 photos this gave me not a side view scan of the face, but a front view scan instead. So i'm just going to combine those two out in my external 3d app.

I am wondering if using markers will help give me an accurate scan or if you could actually manually change the calibration values of the photos themselves, like you would be the one to manually align them. I managed to position some of the misaligned cam planes this way by recalibrating and re aligning photos, but i never could get it in the spot i want since my numerical input in the Fx and Fy values were all just guesswork. I would love to know if there's a numerical map for the entire scene that i can reference to.   

[tincansassoc]

I think you need to buy the professional version to get the advance marking and manual aligning features. However, I believe if you practice taking the photos better using objects that dont move, you may be able to scan a whole face at one time and process the photos in one chunk. Also, the camera you are using makes a difference. Most people using this software are using a DSLR with a professional lens.  Iwas also looking over your images you posted and noticed some of your photos appear square, did you crop these? the EXIF data is useless to photoscan if you crop the photos.

[andy_s]

Hi tincansassoc,

I am still wary of commiting to standard [rather than professional] - and vice versa !

Without wanting to seem lazy[er than I am  ] - could you maybe describe the benefits of 'advance marking and manual aligning features' in a little more detail.

Any comment appreciated. 

[tincansassoc]

Sorry I cannot answer questions about advanced alignment tools, I only have the standard version of photoscan. I have just been doing a lot of reading through these forums about how to improve my own scans.

[andy_s]

No problem tincansassoc, thx anyway...

[Wishgranter]

For head scan a very detailed, its enough with 5 cameras....... but read this http://www.ir-ltd.net/do-you-wanna-live-for-ever and see there for mentioned whitepaper about it......

Hope you get it, a lot of people think that is enough that they put few images and pscan make some magic and its work, but behind it is a LOT of math, and some BASIC about photogrammetry. Try google for it that you understand it and can learn from it for you career.... 

Dont forget that internet-text stuff is not best medium to share feelings, so im not offensive even if it seems to be, if you read my posts.... Im trying to help and direct you, so you understand it better.... 

[Jinny]

Thanks for your input. I'll take note of that next time. I am also using the pro edition and would like to give those marking aligning tools a shot, but don't know how to use them. I thought it would be to mark the photo like place a node in certain parts to mark it with similar parts from other images. As it turns out you can only put a marker inside the scene itself, so i wouldn't know what they could be used for from within a scene. 

Nope i did not crop the photos by the way.

[tezen]

Hey Jinny!
At last you used really small photos at different sizes (down to 3.5megapixels) with a 17mm-lens at ISO 2000 or higher - all this could cause problems. Your photos should be 5mpix (or higher - 8-12mpix is best for speed+details) within 20mm (24mm or 36mm - less lens correction) at ISO800 for less noise (better is ISO100 with good bright studiolights).
Happy scanning!
tezen

[Alexey Pasumansky]

Hello Jinny,

There are some basic recommendations you should follow to get good results (some of them were already mentioned above):

- Since PhotoScan is a photogrammetric kit only original images should be used. Using cropped and transformed images will produce incorrect results. Using original images also allows PhotoScan to derive initial calibration data from EXIF.
- Use maximum possible resolution.
- Photos should have sufficient overlap.
- Try to use as low ISO value as possible. High ISO induces additional noise and therefore reduces the accuracy of the solution.
- In case the lighting conditions are not good for low ISO, use a tripod to get sharp photos.
- Pay attention to the focal depth - photos should provide enough level of detail (recommendation refers mostly to the situations when camera is positioned close to the object of interest).
- Use frame space effectively, in some cases portrait camera orientation is preferred.
- Avoid changing focal length (zooming) during the shooting in case it is not required by any reasons or conditions.
- When taking pictures of the living object keep in mind that object can move (even slight movements causes by breathing, for example, may affect the result). If only one camera is used then siting pose of the model is preferred.

I assume that good results can be achieved without pre-calibrating of the camera if EXIF data is available or some manual work with camera positioning (especially it is almost impossible with single camera case) and markers, in other words PhotoScan Standard edition is fine for the described task.

And also we do not recommend to use cracked versions and do not guarantee any correct work of such versions.

[Jinny]

Thanks for further tips guys.

I just like to know if the alignment features with using markers be of any use to getting a better scan? I still am not quite sure what they are really used for, anyone clarify this for me?

I'm also planning to scan some of the 3dsk photos i have, possibly for study material as well so i can see the 3dimensonal view of what is going on in those 2d photographs. Would it be possible to scan those out successfully, in the very least, decently? So far only one head photo out of the pack have worked as attached in my first post. I was wondering if i could force or manually align some of the unaligned photo projections for the point cloud myself through really understanding the calibration values to position them correctly, since the auto align seems to fail a lot.