The “Butterfly Effect” is part of chaos theory in physics. The theory is that a small change in one state can result in large differences in a later state. The prototypical example is that a butterfly flapping its wings can result in a tornado.
I ran into a “Butterfly Effect” between building the model after Dive #4 and adding the photos from Dive #5.
This post will detail what happened, how I finally resolved it, and my recommendations for others building models of large objects over multiple dives. I put the details of this process and my learnings at the end of the post.
Before I go into a ton of details on the process of building the model after dive 5, here are some screenshots of the current state along with a link to the online model.
Model Screen Captures
The “empty” areas on the starboard side are pretty much the only areas left to do. HOWEVER, there is a problem is that they are covered with kelp that is consistently moving back in forth in the surge which makes aligning photos almost impossible. I have some ideas on how to potentially resolve that, but I’m not sure any of them will work — short of “trimming” the kelp to expose more of the hull.
Sketchfab “Live” Model
Once again, similar to Dive #4, we had some good visibility on the dive. My dive time was 81 minutes and I took about 1,500 photos. My plan was to start at the stern and use the opening in the hull that I shot during Dive #1 as a reference point to align the photos from Dive #5 (upper left photo below).
Other than being confounded by the moving kelp and not finishing those sections, everything went according to plan.
“The Butterfly Effect”
When building the model after adding the almost 1,900 photos from Dive #4, I was a little confused. All the new photos from Dive #4 had aligned but the model only included a small portion of the starboard hull whereas I was pretty sure I had taken photos of more of it. Below are comparative models of version 3 to version 4.
I just chalked it up to thinking I had made more progress on the dive than I actually had — after all, the photos had all aligned.
After adding the almost 1,500 photos from Dive 5, I ran the alignment process and everything aligned — BUT, once again, the sparse cloud looked exactly like it had after Dive #4. What was going on?
I was pretty darn confused. You can see that 6418 out of 6441 photos were aligned but the model looked exactly the same after adding 1,500 photos.
My next troubleshooting step was to align ONLY the photos from Dive #5 to see what that sparse cloud looked like. Almost all the photos aligned and you can see that the stern section of the starboard hull should have been reflected in the model above.
At this point I was starting to question my sanity. How could the photos that were aligned from ONLY Dive #5 that were also aligned into Dive #1-#4 not be in the sparse model?
I decided to post a question to the Metashape discussion forum.
In the meantime, I started to “poke around” to see if I could figure out where all those aligned images had gone since they didn’t get reflected in the sparse cloud. As I was looking around, I noticed a “hairball” near the top of the bridge. The model below shows each camera picture around the wreck. Explaining the light blue vs. dark blue and why they are like that is a bit much for this post but it had to do with the disabled photos that happened as a result of the process I detailed in Dive #4.
WTF was that?!?
I started to “drill down” into it to figure out what was going on and I saw this sparse model that looked a lot like part of the bridge structure but it was like a “reflection” and was oriented in the wrong direction.
It won’t be obvious to the casual observer, but then I also noticed a “mirror image” of the starboard hull structure:
I then started a theory that a misalignment had happened somewhere. The software had “aligned” the photos from Dive #4 but the angle was slightly off at that point and every photo after that point basically folded in on itself.
How to find that? I started going through each dive from Dive #1 and paging through each photo to make sure it was where it should be on the sparse model. I was almost sure it happened on Dive 4 (remember my nagging thought about the model not reflecting my progress).
As I was paging through, I noticed a VERY small change in direction near the top, starboard edge of the bridge and then the “hairball” trailed off! Here are the two sequential photos that resulted in the Butterfly Effect:
In this screenshot, you can see that Image 4623 is highlighted and it is “veering off” to the right. If you trace it back, that “veer” starts between photos 4620->4621 which are the two photos above. The Butterfly Effect!
Fixing The Rats Nest
The next problem is: how to fix it?
I started by deleting out of the model all of the photos after 4620. I also deleted all photos from Dive #5 because all 1,500 of them were “aligned” but in that hairball and the incorrect hull. Next, I re-checked EVERY photo in the sparse model to ensure it was where it should be.
I then started adding chunks of photos from Dive #5 and slowly started to “work backwards” to the problem area thinking that if everything else aligned correctly in these small steps then when it got to that transition, it would be “forced” into correct alignment.
Here are some screenshots that show building the stern area from the existing Dive #1 photos and extending it towards the prop/rudder, etc.
I kept doing this for all of Dive #5, ensuring as each “chunk” was aligned that all the photos were in the right area. I then started “weaving in” all the messed up sections of Dive #4.
Two days after I started trying to align photos from Dive #5, I had “clean” model.
I learned a hard lesson: don’t assume that because photos align that they are aligned “right” and always do a quick, manual sanity check.
I built the final model and it now includes almost 6,000 photos.
Now I’m facing a big problem:
The seaweed itself isn’t a problem per-se, it is the fact that it MOVES during the surge that ALWAYS exists on the Yukon. If I take two pictures of the same exact spot with seaweed in one and not in the other, it will never align. Below is a sequence of photos taken near one of the openings in the hull that shows how the seaweed moves around and can make alignment hard to achieve. Note how the seaweed starts out of the hold, then gets sucked in, then gets spit out again.
I have a few ideas on how to “conquer” this problem but none of them are easy.