Whats going on here?

[lodds]

First time user here. I have downloaded a trial version of Zephyr Lite. My plan is to use Zephyr to export STL file to my 3D printer.
As a trial I took 250 photos of a model and went through the import procedure. All photos bar 1 was imported and processed successfully. However the result looked more like something out of the walking dead which was not expected. Obviously this was due to me mishandling the process but as the wizard was very good cant see what it was i did wrong. I tried to upload the STL file but was not allowed to.

[Andrea Alessi]

Hello and welcome!

it's hard to tell exactly what the cause is without look at the actual photos (if you want to share your dataset with us, you can send them (e.g. via wetransfer or a public dropbox/gdrive link) to support@3dflow.net )

From what I see from your two screenshots, it seems that the cameras are probably oriented correctly (though consider adding a newspaper under the statue for example to help camera orientation) but you have some issues with the surface. With plastic/smooth surfaces this is generally an issue with photogrammetry and can be accentuated especially by bad lighting/noise on the image.

Make sure to get good photos with no direct light, and that your camera settings allow for a sharp image with no noise.

Additionally, you can consider masking the dataset though for this specific case I don't think it's necessary.

[lodds]

Hello and welcome!

it's hard to tell exactly what the cause is without look at the actual photos (if you want to share your dataset with us, you can send them (e.g. via wetransfer or a public dropbox/gdrive link) to support@3dflow.net )

Thank you,
Where can i find the dataset?

[cam3d]

lodds 'Dataset' means the collection of photographs you are inputting into Zephyr :-)

[lodds]

lodds 'Dataset' means the collection of photographs you are inputting into Zephyr :-)

Opps.
just uploaded most of the dataset, Wetransfer has a 2gb limit

Thanks
Tony

[Andrea Alessi]

Hi Tony,

I've had a look at your dataset, unless you forgot to attach the bim files, this isn't been masked.

If you move around the software, like in the cherub dataset ( https://www.3dflow.net/technology/do...ls-3df-zephyr/ ) you do not need to mask the images.

But in your case you have a fixed background, and a moving subject. This creates an incoherent scene. To Make things more difficult, you have a changing background (your TV screen with a show) and very different ligthing across your dataset.

This doesn't mean your dataset is not salvageable (in fact, you already get something out of it).

You can learn more about masking here: https://www.3dflow.net/technology/do...-masking-tool/

I would simply re-shoot the dataset. You can probably get away with about half the photos you took, but 250 is also ok.

Very good you used a tripod, however:

- If you want to use a turntable mode, put a black or white background which will make masking easier and will absorb ligthing in a constant way. Learn more about turntable shooting here https://www.3dflow.net/technology/do...for-turntable/

- Make sure your lighting is constant, if possible.

- Considering that your subject is fairly easy, I would also consider simply moving the object outside, on an overcast day. Then simply move around it and take photos.

- Do not move the subject. If you do (perfectly fine e.g. if you want to reconstruct also the underside) you will need masking for optimal results.

I hope this helps!

[lodds]

That is very helpful, thank you

[lodds]

Just a made a few adjustments and the result is not much better than previous. I purchased a light box and used a white then black background. All photos came out well but the handling of these with the application was far less than perfect IMG 6001 model shoulder was missing and again the surface was full of blemishes. Is it possible to do what I am trying to do and get a flawless file?
I have sent the latest dataset for your consideration.

[Andrea Alessi]

hi lodds, can you send me also the .zep file (and relative _files directory?) so I can see what settings you used?

However, a couple of notes.

- You're still using a turntable shooting and unless you forgot to attach them, the files are not masked. Unless the background is absolutely non-visible, zephyr will pick up features from the background and you'll have an incoherent model with the background. You may want to mask those, or make sure your background is not feeding keypoints.

- smooth surfaces are always difficult, but you should certainly be able to get a good model out of this subject

- consider also adding an "outer loop" where the figure is shown in full body, that can help in orientation if that's what causing issue. Since this object is very complex, I would also consider adding at least one more loop with the camera that is slightly slented from bottom to up, so that you can capture the details "from below"

[Andrea Alessi]

I did a quick test and created less than optimal masks.

This is my result on general/default



it's normal unfortunately you get some bumps in the surface as plastic is difficult to reconstruct due the lack of features (consider using a cross polarizer to minimize reflections)

I've also run a a laplacian smoother filter within zephyr:



You can probably get better results with proper masking, and consider applying the smoothing filter only where needed (as I did it on the full model, it lost some detail on the bottom part that could have easily been maintained).

Eventually, consider fixing the remaining minor issues in blender and reimport the model



This is just a quick result, you can certainly improve it

I'll send you the zep file later so you can at least take a look at it.

[lodds]

Thanks for taking the time to run this. I'll look at the zep file when I get it and try and figure something out.