Loaded images were masked and reconstructed but failed to generate model

[3DEnthusiast]

Hi there,
First off let me say that I am very pleased with my purchase of Zephyr lite Steam edition. This is a very impressive piece of software. I have been using it to create 3d models of static objects that I have come across outdoors when out and about with my DSLR. Up until now I have been shooting handheld and simply walked around the item that I wanted to shoot. This has worked remarkably well for me. Recently I undertook to create a 3D model of a dog skull for an archaeologist friend of mine. This skull only weighs 4 ounces and is a bit fragile so I decided to shoot this item indoors, safe from the wind and rain. I bought some studio lights and set the skull on a home made turntable, which I marked out in 5 and 10 degree increments. My Canon 60D with Canon 28-135mm IS USM zoom was set on my tripod and I shot 72 images at f/11 at 1/60th of a sec. at 75mm using ISO 100.This equates to a 5 degree rotation between each shot taken. The camera was placed 9ft away from the subject. The IS was turned off. The lens has a Hoya UV filter fitted.The images were saved as jpegs. All the images were successfully masked and all 72 loaded and a healthy looking sparse point cloud was produced with every camera view reconstructed. Unfortunately the end result of the 3d model generation produced a garbled result. I have tried several more times to see if the result would come good but have not had any luck. I am obviously doing something wrong but I cannot figure out what. I can send you the dataset.

Thanks in advance
3D Enthusiast

[Andrea Alessi]

Hi 3D Enthusiast,

Thank you for your kind words!

Yes, please send the dataset to support@3dflow.net - it's hard to tell what's wrong without looking at the photos, it's probably easier this way

[3DEnthusiast]

Wow that was quick! That was only three minutes. As soon as I sent my message I realised that you would need to see some photos before any of this will make sense. Unfortunately I now have to rush off to get ready to go to work.Will post the dataset when I get back tonight.

Thanks

[Andrea Alessi]

no problem, happy to help

[cam3d]

Hi 3DEnthusiast - I'm looking at the data-set now and prior to processing it all, one of the first things I noticed was that the subject takes up a very small portion of the frame, so there are very few effective pixels to work with. The metadata suggests that you were shooting at 75mm - I'd suggest zooming in further to fill more of the frame and thus having more pixels to work with! Alternatively, getting the camera closer to the subject would be ideal, but understand that your subject is very small and the lens you are using has a minimum focus distance of 500mm so that may not be possible.

I've just run your data-set with default alignment settings and got a 72/72 but when running the dense there was an error. I suspect this has something to do with the small percentage of the frame actually containing relevant data, and the default dense processing not using 100% of the pixels available. Currently running a new dense with 100% resolution and low speedup to try and resolve this.

Will update shortly!

[cam3d]

Ok so the new settings work fine and got a successful dense cloud - Doing the same with the mesh now and will post a screen shot. - Another thing to note with this dataset is that you only have one elevation captured - By adding an high angle and low angle elevation (bringing the total number of photos in the dataset to 216) you'll have more coverage of the subject and therefore get better results

Mesh was built with 100% resolution as well (normally 100% on any setting is not advised because of processing times vs diminishing returns but this was a unique case) and all other settings default - If you have questions let me know and I'll do my best to answer them!

[cam3d]

Andrea Alessi - This popped up in the log before the dense cloud (partial) - What does it mean specifically?

[16:02:30] Photoconsistency ERROR!

[16:02:30] Qt Warning: QWindowsWindow::setGeometry: Unable to set geometry 500x150+4511+475 on QWidgetWindow/'ProgressBarQtClassWindow'. Resulting geometry: 506x167+4511+475 (frame: 8, 31, 8, 8, custom margin: 0, 0, 0, 0, minimum size: 500x150, maximum size: 500x150). -

[3DEnthusiast]

Good morning cam3d,

Thank you so much for your perusal of my dataset. I should first confess that I am in no way an an expert photographer and only recently upgraded to a DSLR from a digital compact super zoom camera. Photography interests me but I am not very knowledgeable as yet. I have relied almost totally on the fact that modern cameras do most of the work for you. Several years ago I did buy a film SLR but this was rapidly turned into a expensive paperweight by the collapse of film and the resulting increase in the cost and difficulty of getting a reel of film developed. This pushed me back towards digital and I have stayed there ever since.
The last few weeks have been spent trying to figure out f/stops, depth of field, aperture priority, the inverse square law, acceptable shutter speeds, and so on and so forth. A lot of my spare time has been spent recently on the depth of field simulator.com website in an attempt to come up with a workable focal length, combined with a f/number and distance from subject that would give me decent depth of field. My first run of shots was indeed at 135 mm close up so as to maximise the amount of picture in the frame but the depth of field was too shallow so I backed up and kept experimenting with focal length, f/stops, lighting setups and so forth. This eventually led me to 75mm at f/11 at 9ft as this seemed an acceptable compromise. It was only then that I felt ready to commit a run of photos to Zephyr. Three elevations were shot as I was planning to merge these together to create a seamless 360 degree model, but as these kept failing I eventually limited myself to shooting just one elevation and running that through Zephyr until I could solve the problem of garbled results.
The dog skull is 7 inches long by 4 inches wide and stands 3 inches tall. The decision to go with 100 % resolution can be explained by my desire to get the most resolution out of my model and my overall ignorance of what an appropriate setting would be.

[Andrea Alessi]

Andrea Alessi - This popped up in the log before the dense cloud (partial) - What does it mean specifically?

[16:02:30] Photoconsistency ERROR!

[16:02:30] Qt Warning: QWindowsWindow::setGeometry: Unable to set geometry 500x150+4511+475 on QWidgetWindow/'ProgressBarQtClassWindow'. Resulting geometry: 506x167+4511+475 (frame: 8, 31, 8, 8, custom margin: 0, 0, 0, 0, minimum size: 500x150, maximum size: 500x150). -

Hi Cam,

the first line tells you that the photoconsitency faild. My guess is that you raised the resolution too much and there was not enough video memory for the photoconsistency. I would need the rest of the log to investigate further.

The second one can be ignored, it's a warning on the size of the window - not sure why it happens, probably some leftover code that does nothing, but i wouldn't worry about it - we'll have a look at it though.

[cam3d]

Andrea Alessi - Higher resolutions were successful, but lower not so much. Log + project have been sent through.

3DEnthusiast - Great to hear you're learning about all the fun elements which come together to make a photo. Andrea is going to be able to give some more specific feedback once he as seen the logs, there might be a rare bug that is making things extra hard. As for camera settings and framing, based on the information you have provided I'd be inclined to use settings as follows:

600mm from camera sensor to subject
50mm focal length
f/11
Shutterspeed - Whatever you need as long as you're using a tripod (also a shutter release cable or 2sec timer will reduce the chance of any motion blur cause by your hands)

To be fair though, without a more hands-on approach I am just making guesses based on my own experience and this may or may not be applicable - One thing to add is that even if the whole subject is not perfectly in focus, it's not a deal-breaker as long as the out of focus surface is in focus in other shots.

One other thing I'd like to clarify: When you say garbled, can you please post a screen shot of the results your getting? There are some other things which might be causing problems but it's very hard to tell without the visuals

[Andrea Alessi]

The failed photoconsistency problem reported by Cam is actually a bug in the photoconsistency process, the fix is underway - thank you for sharing the dataset!

[3DEnthusiast]

You are more than welcome. It may help someone else out in the future. A new dataset of 71 images was captured this morning using the suggestions provided by your good selves and masked and sent down the Zephyr pipeline. This produced a very believable representation of a dog skull. This was shot at 5ft from the subject at 135 mm and staying with ISO 100 and f/11. The dataset and the resulting .zep file have been sent to Andrea via email and OneDrive.