But How Much is it Worth?: How 3D Printing to Shows Why Archaeologists Don’t Understand This Question

This is a repost of an entry written for the Virtual Curation Laboratory‘s blog and was the foundation for a presentation at the Southeastern Archaeology Conference’s (SEAC) 2016 lightning round.

Many of us have heard about archaeologists modeling and printing artifacts in 3D. There certainly are valuable research applications for this technology, but for me the really exciting prospect is in outreach and education. Sometimes however, I get the impression from others that they think this is neat, but a little gimmicky. I think that this perspective is understandable, if a bit short sighted.

3dvcu_100_5000
This bone toothbrush (right) is far too fragile for it to be handled frequently. (Photo by Bernard Means)

There are a lot of things we can do with a 3D model (many of which you may have read about on this blog) that you cannot do with the real artifact. We can let people hold artifacts that are normally too fragile, put magnets in the edges of a broken artifact for people to reassemble (my personal mission), even make tangible displays for those who are visually impaired.

img_1719
This is a hand forged nail, it is 2 1/2 inches long and was probably made sometime before 1900. That is all it can possibly tell us. (Photo by author)

While each of these are exciting, what really gets me worked up is the ability of 3D models and prints to pull the focus away from the value of the object and move it towards the value of what the object teaches us. When at a local heritage day, we often bring along a case of artifacts that were donated by collectors, so we don’t know where they came from. These are wonderful because we can let people touch them and not worry overly about them getting damaged. Unfortunately, when people ask me about these artifacts, all I can typically do is tell them what it is and maybe how old it is. Inadvertently, this is doing the exact opposite of what we want. It is reinforcing the idea that this artifact is valuable in and of itself.

However, 3D printing allows us to present an object that was recovered archaeologically, even if it is not real, and then focus on what that object taught us. For example, I have three printed, refitting artillery shell fragments from a Civil War site here in Florida. I can tell an audience that we know they are from a shell that broke up before it exploded due to the size of the fragments and the fact that the fuse is still attached. Then I can tell them that these pieces made a straight line pointing right at where the Confederate artillery was supposed to be! We know this because of how we found and mapped these artifacts.


And they
get it. In my experience, I can actually see them understanding this point in a way that I rarely achieved by using an abstract example or “what ifs.” This is the object, or a representation of that object, and here is what we learned because of how we recovered it. If you want to teach a preservation message, highlight the difference between archaeology and collecting, or just get the visitor to walk away understanding what archaeologists do, I have yet to find a more concise and consistent way of achieving this.

This is not to say that these 3D models will replace the artifacts. There is still a benefit to letting people see and interact with the real thing. However, as 3D printing becomes more accessible I fully intend to include at least one example at heritage events that shows the value of what we learn from artifacts.


Tristan Harrenstein is trained as an archaeologist and has a passion for outreach and education. This is fortunate as he is also an employee of the Florida Public Archaeology Network, an organization dedicated to promoting the preservation and appreciation of our archaeological resources. He shares a blog space with his boss (Barbara Clark) and you can read more blog posts on other archaeology related subjects that tickle our fancy here.

The Basic PhotoScan Process – Step 4

Photogrammetry has incredible potential in archaeological research and education. However, despite Agisoft PhotoScan’s relatively simple initial workflow, things get complicated pretty quickly. Those of us using the program tend to learn by solving problems as they occur, but this is a very piecemeal, time-consuming, and often frustrating process. Currently, anyone getting started with the program must either go through the same thing, or find someone to offer guidance.

In this series I will assemble all the separate tips that I have learned or found into a step-by-step guide on the basic process (posted weekly). I do not consider myself an expert in PhotoScan. If you are familiar with the program and have any corrections or additions, please let me know. Each week, the previous step will be edited to include any comments and placed under the “Resources” menu to serve as a guide for beginners.

The previous steps can be found here.

Step 4: Cleaning the Sparse Point Cloud

Once processing is complete, you will have what is called a sparse point cloud. This is essentially a rough sketch before you ink the lines in during Step 5.

Hopefully, you will be able to make out the shape of your subject, though there will be a lot of points you either do not want, or do not make sense. If you hit the camera icon on your toolbar some blue squares will pop up with the name of your photos attached to each. These are your camera positions in relation the the subject when you took that photo. Pretty neat!

Checking Alignment

photo-alignment
The photo in the top-right did not align and is not being used by the program. (Image by author)

Before proceeding with the cleanup, scroll through your photos and make sure there is a green check mark in the top-right corner of each. This mark tells us that PhotoScan successfully aligned that picture (at least it thinks it did). If a photo failed to align then the program could not figure out where it was supposed to go. This is almost always a result of an out of focus picture, bad lighting, or not enough overlap between photos.

To fix this try right-clicking the offending photo, select “Reset Camera Alignment,” right click it again, and select “Align Selected Cameras.” This is a long shot, but it does work occasionally. If the photo still will not align, you can try messing around with placing markers, but you might be better off just removing or retaking the photos.

Gradual Selection

The first thing you should do to clean up your model is to use the “Gradual Selection” tool. This process is pulled directly from dinsaurpaleo’s blog and boy does it make a big difference. His description is a little hard to follow, so I will include it here again.

  1. Critical! Right click your “Chunk” on the left and select “Duplicate Chunk.” Make sure the “Copy of Chunk” is in bold text before proceeding (this tells you which chunk you are working on).
    copy-of-chunk

This way you will keep the original sparse cloud unaltered in case something goes wrong.

2. Under the “Edit” drop-down menu, select “Gradual Selection…” and a window will appear.

3. In the window, click the drop-down menu next to “Criterion” and select
“Reconstruction Uncertainty”

4. Next to “Level” enter “10” and select “Okay.”gradual-selection-window

You will see that a lot of the points in your sparse cloud turned pink. That means they are selected.

5. Hit your “Delete” key and those pink dots will disappear. (Yes, I’m serious. You won’t be sorry!)

gs-hack
You are better off without all those points. (Image by author)

6. Repeat numbers 2-5 one more time.

optomize-camera7. Open your “Tools” drop-down menu and select “Optimize Cameras…” and a window will pop up.

8. Select all checkboxes except for the last two (I admit to having no idea what these really do) and click the “Okay” button.

This will take a few minutes to reset your photos without the inaccurate points you just deleted. If you deleted a lot of points you might get a “Some cameras have insufficient number of projections and will be reset. Continue?” If you get this popup, click “yes” and the program will try to reset the photos based on the remaining points.

9. Now, go back to your “Gradual Selection…” window and the “Reproduction Error” option should already be selected next to “Criterion.”

10. If the slider total is less than “1” you can skip to number 15.

11. Otherwise, set the “Level” to “1” and select “Okay.”

12. Hit your “Delete” key to delete the selected points.

13. Open your “Tools” drop-down menu again and select “Optimize Cameras.”

14. Your previous setting should still be selected. Click the “Okay” button.

15. Go back to your “Gradual Selection…” window one more time and choose the “Projection Accuracy” option next to “Criterion.”

16. Play around with the number until about 10% of your points are selected and click “Okay.”

17. Hit your “Delete” key.

And you are done! Hopefully, your sparse cloud looks a lot more like the object you were trying to model. I have, occasionally, had this process be too aggressive with a weak sparse point cloud, resulting in big holes in the next step. That is why we made a duplicate!

All that is left to do is some last tidying up. Spin your model around and use your “Freeform Selection Tool” on your toolbar to select and delete any leftover points that you do not want to model. This includes background points and the inevitable nonsense points floating around your model.

freeform-tool-and-resize-tool
The “Freeform” (red), “Resize” (blue), and the “Rotate” (green) tools. Useful Tip: Quickly swap between these tools and your cursor with the spacebar. (Image by author)

Once everything is cleaned up to your satisfaction, use the “Resize Region” and “Rotate Region” tools to manipulate the box surrounding you subject until it is just bigger than what you want to model. This reduces the area PhotoScan has to process and speeds everything up.

Step 4 is done! This part is by far the most involved because it will define the quality of the rest of your model. Things are simpler from here on out.

The Basic PhotoScan Process – Step 3

Photogrammetry has incredible potential in archaeological research and education. However, despite Agisoft PhotoScan’s relatively simple initial workflow, things get complicated pretty quickly. Those of us using the program tend to learn by solving problems as they occur, but this is a very piecemeal, time-consuming, and often frustrating process. Currently, anyone getting started with the program must either go through the same thing, or find someone to offer guidance.

In this series I will assemble all the separate tips that I have learned or found into a step-by-step guide on the basic process (posted weekly). I do not consider myself an expert in PhotoScan. If you are familiar with the program and have any corrections or additions, please let me know. Each week, the previous step will be edited to include any comments and placed under the “Resources” menu to serve as a guide for beginners.

The previous steps can be found here.

Step 3: Align Photos

Now that PhotoScan knows which pictures you want it to work with, it is time to line them up spatially in relation to each other. Fortunately, the program does all the work here.

Once again, go to your “Workflow” menu and this time select “Align Photos.” A window will pop up (pictured below) with some settings to play with. If you want a really thorough explanation for what each does I recommend checking out: http://www.agisoft.com/forum/index.php?topic=3559.0.

align-photos-windowIn a nutshell though:

  • “Accuracy” is exactly what it sounds like. Higher will get you better results, but at the cost of a longer processing time and a model that is more taxing on your computer. I use “Medium” or “High” as the “Highest” setting takes a very long time on my computer and I am not sure it makes that big of a difference.
  • “Pair Preselection” is set to “Disabled” by default. Change this option to “Generic” to greatly speed up the processing time.

If you click on the “Advanced” tab you will get several more options.

  • “Key Point Limit” describes the maximum number of points the program will try to draw from a photo. A higher setting will improve camera alignment and increase processing times. The default is 40,000 and the post above was unable to see a difference with anything above that point. Because I have trouble with the idea of letting go of accuracy, I keep this set at 1,000,000.
  • “Tie Point Limit” sets how many of these points the program will use to align the photos to decrease processing time. If you set this to 10,000 the program will only use the ten thousand best points from whatever you set the “Key Point Limit” to. To use all the points, set this to “0.” If you feel that the photos are taking too long to align, then it might be worth adjusting this setting.
  • “Constrain Features by Mask” is a setting you will probably only need if you used a turntable for your pictures and have applied a mask to your photos. Leave this unchecked otherwise.
  • “Adaptive Model Fitting” improves your camera alignment in ways I do not understand. Leave this checked.

Once everything is set, hit “Okay” and let PhotoScan do its thing. Depending on your settings and how many pictures it has to process, this could take anywhere from a few minutes to hours.

The Basic PhotoScan Process – Step 2

Photogrammetry has incredible potential in archaeological research and education. However, despite Agisoft PhotoScan’s relatively simple initial workflow, things get complicated pretty quickly. Those of us using the program tend to learn by solving problems as they occur, but this is a very piecemeal, time-consuming, and often frustrating process. Currently, anyone getting started with the program must either go through the same thing, or find someone to offer guidance.

In this series I will assemble all the separate tips that I have learned or found into a step-by-step guide on the basic process (posted weekly). I do not consider myself an expert in PhotoScan. If you are familiar with the program and have any corrections or additions, please let me know. Each week, the previous step will be edited to include any comments and placed under the “Resources” menu to serve as a guide for beginners.

The previous steps can be found here.

Step 2: Add Photos

There is no model until you tell the program which photos to use. While in PhotoScan, simply select the “Workflow” dropdown menu and click on “Add Photos” or “Add Folder.” Navigate to your photos, select them, and click “Open.” That is it!

Okay, not quite. One way to make sure your photos align as best as possible, and to increase the detail of your model, is to use a program like Camera Raw. All you need to do (if using Camera Raw) is open all of your pictures together and increase the contrast to just before the shadows start merging together. Then you save these versions with the exact same names in a separate folder, do not save over your originals! If you follow this step, add these high contrast pictures instead of the originals.

contrast-example
Flat lighting (left) is has some hard to see details. Slightly adjusting the contrast (middle) can help this texture pop out for PhotoScan. Don’t take it too far though or you are losing details (right). (Images by Author)

That is all for this week. Look for step three next week when things really start to get complicated!

The Basic PhotoScan Process – Step 1

Photogrammetry has incredible potential in archaeological research and education. However, despite Agisoft PhotoScan’s relatively simple initial workflow, things get complicated pretty quickly. Those of us using the program tend to learn by solving problems as they occur, but this is a very piecemeal, time-consuming, and often frustrating process. Currently, anyone getting started with the program must either go through the same thing, or find someone to offer guidance.

In this series I will assemble all the separate tips that I have learned or found into a step-by-step guide on the basic process (posted weekly). I do not consider myself an expert in PhotoScan. If you are familiar with the program and have any corrections or additions, please let me know. Each week, the previous step will be edited to include any comments and placed under the “Resources” menu to serve as a guide for beginners.

Step 1: Taking Photos

Making a model in PhotoScan works like a reverse panoramic. Instead of taking photos from a point, you are taking them of a point. Since everything that follows is based on these photos, this step is absolutely critical. If you do not get this right, you are setting yourself up for a giant headache and/or a second photo session.

The Pattern

Similar to a program that stitches pictures together for a panoramic image, PhotoScan needs at least a 30% overlap between your pictures to make sure it can figure out their orientation. When photographing an object like a headstone I like to do one ring at a high angle, one at the middle and one at ground level pointing up. Also, do a series of shots linking the high angle to the low. This will help to ensure PhotoScan can connect each pass together.

cw-marker-with-photo-alignment
Each blue square represents a picture that PhotoScan has aligned. This model did not need the fourth pass, but it did need better coverage of the top. You can see the results yourself here. (Photo by author)

Camera

A high quality camera, and the skill to go with it, will allow for higher quality models and final textures. However, you can get some excellent models from point-and-shoot cameras (like in the above image) or even phones! Which you will use depends on what is available, the conditions, and the purpose of the model.

Format

If your camera allows it, I recommend taking photos in the RAW format as this is uncompressed and lossless (otherwise JPEG works). PhotoScan cannot use RAW images directly, so you will then need to convert them to TIFF files before you can import them. I use Camera RAW for this, though I am sure there are other programs you can use.

Lighting

parish-cemetery-headstone
Due to bad lighting, I was never able to get the images for this headstone to align correctly (Photo by author)

PhotoScan does not deal well with sharp changes in lighting or white surfaces that are too bright. You will save yourself some frustration by making sure your lighting is as even and flat as possible. This means that photographer lights will help indoors. If modeling outdoors, take your photos around sunrise/sunset, with cloud cover, or using tents or tarps.

Focus!

Having your photos in sharp focus is critical to a successful model. Any picture where your subject is blurry should be deleted as it will only cause problems. A monopod or tripod will help you avoid this issue to begin with, especially when photographing small objects up close.

blurry-stump-1
Nothing good will come from blurry pictures like this one. (Photo by author)

Depth of Field

This refers to how far your focus goes beyond the focal point. I will let you look up further discussion about this, but essentially, if your depth of field is too narrow only part of your subject will be in focus, if it is too broad then the background will also be in focus. Neither of these will necessarily kill a model (unless working with a fixed camera position), but they will at least result in unaligned photos, or lots of extra points from the background.

Number

This will depend on the size of your subject and the level of detail you are going for. More photos will allow for more detail to a point, though there are diminishing returns. I typically shoot for around 70-80 as a standard, though I have made some good models with as few as 40 pictures.

The key is to make sure each photo has the needed minimum 30% overlap with the previous one. You will learn how many you need with some experience. Do remember though, 1,000 photos will not do you any good if your pictures are bad.

 

That is it for step one. Please weigh in if you have anything to add! Next week will be a short post about adding the photos to PhotoScan.