URE student led 3D Project


(Uremuseum) #1

Dear Heritage Enthusiasts,

At Reading University both the Archaeology and Classics departments have a wide range of teaching collections, ranging from Greek and Roman ceramics, to early prehistoric tools. Students are able to access this material due specific workshops or classes, but out of class acess is restricted due to the rarity or fragility of objects. Our student lead project will focus on developing a prototype virtual teaching collection, accessible via sketchfab. As this is a student based project we are relatively new in terms of 3D reconstruction and any feedback/comments will help guide future development.

the link below shows a fully fleshed out model

Attic Black Figure Lekythos by The Ure Museum on Sketchfab

Thanks for your time,

URE Museum (James and Dan)


(Uremuseum) #2

Models created using a combination of Photogrammetry (Memento Beta)
and Next Engine 3D scanner (ScanStudio)


(Nebulousflynn) #3

Looking good!

The annotations really add value to 3D scans and it looks like you are building up quite the collection :smile:

My only (hopefully constructive) criticism would be that the quality of the texure is quite low. If you have even a mid-range compact camera (12 mega pixels or so, Sony/Canon etc, wide angle lens) you will pobably get better (higher quality) results from Memento. If the univesity has a budget for this student pogram, they might like to consider buying Agisoft's PhotoScan or similar.

Once you upload your models to Sketchfab, play around with the 3D settings - try PBR rendering or tweak the lights and post-processing filters. A made a short tutoial video that you might ind useful too: https://youtu.be/hXEPvaL791o

I just followed you and look forward to seeing you uture work in my feed!


(Uremuseum) #4

Hi Flynn,

Just the man we wanted to hear from, thanks for your amazing feedback!

We are currently using a students Cannon 550D, due to budget restrictions etc, I shot most of the images on RAW, for the Medusa, Lykethos and the Poppy flask, the quality may have suffered on later models due to time constraints etc. (less image processing) I have a trial version of the Agiosoft program and shall attempt to produce a model with it in the future, although processing takes a while on a laptop :joy:

Shall experiment with the filters now,

Thank you again :smile:


(Ben Kreunen) #5

I'm close to moth-balling our NextEngine scanner.. happy to elaborate why but short answer is I can drive Photoscan as fast as the NextEngine for an end to end process and get much better textures. If the educational price of Photoscan is still an issue, you can do pretty much most of what you'd need to with the standard version. Include a scale bar in the photos and you can correct the scaling in Meshlab at the end of the process.

Other useful tools (free stuff):
xNormal - generate a normal map from high res and low res version of the mesh to add more detail in Sketchfab.
MeshMixer - relatively easy to use. Good for editing mesh artefacts (small spikes, holes etc...)
And of course you can access a whole range of awesome Autodesk software. http://www.autodesk.com/education/free-software/all e.g. Adding scale bars in 3DS Max... or Blender

Having a good photography space and lighting can make a big difference.


(Nebulousflynn) #6

All good advice right there!

FWIW and just to compare gear/workflows, here's my current set up:

  • :camera: Canon G7x camera, hand held, shooting JPGs, full auto, around 100 - 200 images avg.
  • :cd: Agisoft PhotoScan Standard Edtn., medium settings (can share more no request).
  • :cd: Blender 3D for a bit of editing / retouching meshes
  • :cd: Photoshop for some texture edits if needed
  • :wrench: Sketchfab set up pretty much as in my video
  • :computer: for the nerd in you: I'm using a 2011 Macbook Pro, 2.2Ghz i7, 16gb RAM, AMD Radeon HD 6750M 1024Mb (yes it gets pretty hot during processing! :fire: )

Obviously this is not a definitive set up and and would change depending on your needs. Bear in mind that I generally scan for online presentation (i.e. on Sketchfab) and not with 3D printing or scientific research in mind.

Here's my latest scan with above setup, scanned outside on a cloudy day:

Fountainhead in the shape of a Faun by Thomas Flynn on Sketchfab


(Uremuseum) #7

We have been experiencing issues regarding texture on certain objects too, and again I much prefer photogrammetry. I guess I shall have to dive into some photoscan tutorials :smile:

thanks for your help,

Dan


(Uremuseum) #8

Quick Question,

Is agiosoft temperamental when it comes to camera settings?
as using Memento I have found that having a higher f/stop allows the program to better connect the images due to the fact that the background is also clear, or with agiosoft is it just focusing on the pixels variation on the object?

Many Thanks,

Dan


(Ben Kreunen) #9

It deson't distinguish between background and object. Image is broken down into small patterns. each pattern equates to a point. After that it's just math on points. Image quality affects the accuracy, reduces false positives (eg. High ISO noise) and thus improves definition between object and background. Masking the images can help in some situations. Increasing number the number of images can also be a big help.


(Nestor F. Marques) #10

Hey there @uremuseum,

Seems like the kind of piece suitable for photogrammetry but not for a 3D scanner. Your results with photoscan are not so bad... maybe masking a little better the images will improve it, in the upper part. Seems like the photos taken by the scanner are not the best for this purpose, so using the RAW ones will always make it better. Can you post one of the images here to see if anyone can open it?

In any case I would recommend you to retake the images with the best camera you can get because almost any camera will be better than the one that is mounted in a 3D scanner. Also... having the object in a more neutral environment will make masking easier in photoscan afterwards.

Just keep trying! You are on the right track!


Share Gear / Software or Hardware Comparisons / Scan Experiments
(Uremuseum) #11

Amazing Models and feedback!
thank you very much, once I have a bit more free time I shall try refining the process,
and maybe convince the museum to buy Agiosoft :smile:
that lutrophoros is just mind blowing :smiley:


(3dsam79) #12

personally i have found that this depends on whether you are taking pictures as you move around the object, or rotating the object on a turntable and keeping the camera static

Having processed several image sets in both Photoscan and Memento, i found that Photoscan often struggled with turntable shoots, adding an annoying outline to the point cloud the same color as the background. Using a much larger textured turntable, ensuring that it was visible in all shots even the low ones, greatly improved the success rate and the results.

I mainly use memento because the meshes are more detailed and its very simple to use. Plus given the way i work, i like how the processing is done on their servers rather than my computer, meaning i can work on other stuff while waiting for it to process:)
Photoscan on the other hand does produce higher quality textures and can handle more than 250 images (Memento's limit). it does take a while to process though and requires a fairly beefy computer, they recommend 128gb of RAM on their site but you can still process nice results on medium setting with 16gb of RAM.


(Uremuseum) #13

I have to agree of the much more 'user friendly' feel to memento and the remote processing is a life saver as I operate on my laptop with lots of other work to be doing! :smile:
I like the idea of using a turntable and after watching @nebulousflynn tutorials I shall have a practise as reducing the amount of crouching and movement will make the process a little less of an exhausting procedure :stuck_out_tongue:
I love the detail on your Green Man model


(Nestor F. Marques) #14

Thanks for the compliments @uremuseum!

Regarding the turning table thing, both the bronze head and the lutrophoros have been scanned using this technique in Photoscan. Usually it's the best method for small objects, and for me, Photoscan does a very good job with them.

I understand Memento is more user friendly and more importantly "regular computers friendly", but I don't like it as much because of the automatization process. It doesn't allow you to take care of the process and, as a result of that, a great capture usually comes out much better in Photoscan (if you know the right buttons to touch :wink:)

@uremuseum I recommend you to buy a Photoscan license (there are some good educational and institutional discounts you can apply for) and give it a try.


(Uremuseum) #15

Hi all,

Because we have had so much trouble with our NextEngine scanner, I have decided to try and repurpose the photos the scanner took for textures for photogrammetry. Initial tests with Photoscan haven't been too disastrous, but they are not great either! The reason for this is because the jpgs from the scanner include a mask that has been applied (and because the photos aren't the best for photogrammetry in the first place). However, I am hoping that the raw files do not contain this mask, and could be used to make better models. The problem is, I cannot open the raw files!

If anyone has any advise as to how I might be able to open them, that would be amazing! (Photoshop has the latest version of Camer RAW, so I assume the just don't have NextEngine raw on there...)

Thanks!

James


(Nestor F. Marques) #16

Well... I just replied... but seems like my post didn't publish. So I'll write it again, but maybe later it appears twice!

I was telling you that you've got pretty decent results with photogrammetry, taking into account that the images you used are the ones from the 3D scanner (which tend to be very low quality most of the time). Make sure you use the RAW ones for better results.

Upload one of them to see if any o us can open it if you want. Otherwise I recommend you to retake the images with a regular DSLR, almost every camera will have the same or better results than the 3D scanner sensor (which usually is not designed for geometric reconstruction, but only medium quality texture)

Just keep trying, you're on the right track!


(Ben Kreunen) #17

The RAW files AFAIK are the raw scan data for each scan and not images. I've never seen that kind of masking on the jpegs, curious. That's from the source files and not the OBJ export?