Trying to scan a small item

[arcturus]

Good evening,

It seems that the standard introduction to this forum is to say that one is new to Zephyr. That is true, put aside the fact that I have been trying to scan a small object for the last three weeks without any success (not at all, nothing, nada, zero !).

I read a lot and finally found the forensinc case study which looks pretty much like what I am trying to do. I copied their targets as they looked more clever than mine, but again with no avail : Zephyr ignores my item. It reproduces correcty the surroundings, but not the item.

I usually shoot 24 images with a K-5 IIs and a 100mm lens at 100 ISO and F32 on a tripod. I produce 16Mo jpgs without compression, strait from the camera, no LightRoom, nothing. Typical image appended. I understood that my item texture is quite plain, and tried to improve it by adding stips on it, but it did not work either.

What am I doing wrong ?

Thanks by advance,

Regards,

[Andrea Alessi]

Welcome!

unfortunately your subject isn't very textured - this will be your main challenge.

use the markers only if you need them as a reference for distance constraints / rescaling. The most common misconception is that they help in the SfM phase - they don't. They're just to have zephyr automatically grab a control point.

To help the SfM phase, you can consider putting soemthing very textured on the table, e.g. a newspaper. Keep in mind that the subject must also never move (or you'll need to mask the subject, making any aid such as the newspaper useless).

Consider dirtying the subject if the texture isn't necessary and you just need the geometry. Otherwise, consider an approach with a projector ( see here at 4:20 https://www.youtube.com/watch?v=E06kgYBftak ) to add temporary features and use two shots.

Also consider framing more of your subject, as right now basically 8/9 of the image are "wasted" as you're framing stuff you're not interested in reconstructing. Consider doing a loop like so, but then zoom in more.

Feel free to share your dataset with support@3dflow.net as it's difficult to give you more tailored advice without looking at the photos.

24 images are also a bit on the low end. Consider taking at least a couple of orbits of the subject at different heights for example. You want as much parallax as possible to help with the geometric reconstruction.

[arcturus]

Thank you very much indeed !

Good news is that you do not tell me it's not possible....

I made the confusion between markers for dimension measurement and neccessary texture for the software to have something "to grab on".

Using a projector : I may have a of practical realisation problem, but is it possible with the free version, and where do I find the relevant command discribed in the manual ? Alternatively, I have the option to basically paint zebras on the piece. Would I need them on a regular or random pattern ? I could also use a very mat varnish, what would be better ?

Framing more the subject : I may have a problem of physics : the closer to the subject, the smaller the DOF... f/32 is the best I have. It gives a theoretical 8mm DOF for a 30 cm distance from the subject, which is more or less what I did (my lens is a 100mm macro), even if the theoretical calculation seems pessimistic. I'll try as suggested on a printed page.

Number of images : I shot 8 images per orbit, that's one every 45 degrees, and shot 3 orbits, one 10-15 degrees from the horizontal, one at 45 degrees (take or give) and a last one at 70-75 degrees. Is there an optimum between the number of orbits and the number of images per orbit ?

I'll keep you posted.

Thanks again,

Regards,

[cam3d]

Hey arcturus - To answer your questions:

- If you only need the mesh surface, you can just include the projector overlay in the image set, and treat as normal.
- Better than a projector, in my opinion, is adding lots of high contrast features, like finely ground pepper for dark features, or talcum powder for light features.
- Yes physics is an issue, always ;-), but I must add here that you need a lot of light to compensate for high f-stop. I assume you're just pushing the shutter speed out to long times to compensate, and this can, at small scales, introduce vibrational error.
- Lots more images! I recommend at least 28 per rotation (every 12 degrees) or higher. More images = more context = better results.
- This tutorial is pretty good if you haven't read it already: https://www.3dflow.net/technology/do...for-turntable/

[arcturus]

Thank you cam3d,

You are right for the f-stop, I am shooting at speeds in the area of 15-20s. I was not aware of potential vibrational error. I just switch off the camera stabilisation and set a shutter delay.

I have been very interested by tutorial 12, don't know how I missed it. Regarding LightRoom : is it possible to use the sharpness enhancement tools on top of the normal lighting tools ?

Similarly, but I suspect the answer is no : is it possible to use image stacking ?

Andrea Alessi , I obtained something (not usable, but existing) with a closer framing over a printed sheet of paper. Lighting conditions were not so good, printing was done with an inkjet printer on white paper, but I had enough DOF. I will try to find something printed white on black (tutorial 12 say black background for light objects), with a proper sharpness (my inket printer is not sharp enough...).

Regards,

[cam3d]

arcturus - I'd avoid any sharpening or other pre-processing of images other than converting them from RAW > JPEG prior to processing in Zephyr. Image stacking is likely to introduce more variables and more reconstruction issues.