HMCS Yukon (DDE 263) Photogrammetry Model

FINALLY!

It seems apt that this is being published on “Groundhog Day” since it seems like I kept waking up in a loop and diving the same-ole Yukon trying to finish my same-ole model.

The model isn’t perfect and I faced many, many challenges when trying to build it, but I have what I consider to be a passable model of the HMCS Yukon wreck. Below are details of the process along with some very interesting statistics and how this project unfolded.

Online Model

Before I get to all of the details, here are links to the on-line 3D models:

Mobile Version (800k Triangles)
Desktop Version (8M Triangles)

The model I built in Metashape is 150 million faces so I had to decimate the model quite a bit for Sketchfab.

Note that there are a couple “easter eggs” in this model. One is very large but mostly hidden and the other is small and also a little hidden but can be found. It is probably easiest to find it on the desktop version since it has more details. Please reach out if you find the small one. 🙂

Model Screenshots

Below are some screenshots from the model:

Model Details

This was a “master class” in resolving alignment issues. I must have tried 15-20 different ways to get all the photos to align. I had two primary ways that I manipulated the process in an effort to find one that worked correctly. The first was to alter the order in which I added the photos from dives. The second was to pick groups of photos from each dive that I thought would align. Just for historical purposes, the final order for most of the photos was Dive 10 -> Dive 11 -> Dive 3 -> Dive 2 -> Dive5 -> Dive 4.

The primary reason for that order is that I knew that Dive 10 and Dive 11 were the key to getting the starboard hull correct. I then started using different orders for the dives with “good” photos.

For example, here is what happened after I aligned Dive 11 -> Dive 1 -> Dive 2 -> Dive 5:

I had a particularly hard time at the stern guns. Without going into a ton of details, the “problem” is the software has an “stronger affinity” to photos from the same dive if you don’t turn off “pre-select” and if you turn off “pre-select” then alignment time goes up significantly for projects with a lot of photos. Here is a screen capture that shows how the alignment can be slightly off at the stern guns:

I didn’t want to just let that go so it took a LONG time to get it right. The cargo area in the stern is still not aligned correctly and the mount near the bow gun has a “duplicate” but it will have to do!

All told, I probably spent two weeks on-and-off playing with order and photos to get them to align.

Once I got them to align, then came the process of building the model and the texture. This is where things get really interesting in terms of processing time. For Christmas, Santa Claus got me a new Mac Studio computer with 64GB of memory and an Apple M2 Max chip that has 12 core CPU and a 30-core GPU. The software makes good use of GPUs for most of the processes. In all, it is a pretty darn powerful computer.

Here are some key stats:

• The model ended up being 11,750 photos (I shot over 15,000 photos in total)
• Depth map creation for High Quality, Mild Filtering took 16.2 hours
• Mesh creation took 9.25 hours
• The final model has 150 million faces
• The high quality 8k x 8k texture took 26 hours

Therefore, not including alignment time, the model took 51.5 hours to build!

Note that due to the complexities of getting things to align, there are probably 2x the number of photos I should need for the level of detail in this model but “it is what it is.”

Eleven Dives to Complete

Normally, a wreck this big would take me 5-6 dives to complete (if the conditions were always good) with the level of detail I want. I base this upon the 4 dives it took me to make the USS Moray model. The Moray is shorter by about 50 feet and the beam is less and there is significantly less detail. It turned out to be about 4,000 photos.

However, this project took me 11 dives over 14 months to complete. There were three big challenges to this project:

1. The conditions are highly variable and things are constantly growing and changing on the wreck (especially over a 14 month time frame)
2. The seaweed on the starboard side hull that constantly moved and hence caused pictures not to align. I tried many different techniques to work around this (details below). The methods included running a cave line and adding cave cookies, having diver “hold on” to the wreck as a reference point, taking photos from far above the wreck, etc.
3. The wreck changes over time. You can see parts of this wreck literally moving in the surge and the wreck shifts so that photos taken months apart won’t align

Below are two pictures of approximately the same area on the ship on two different days. Same camera, same lens, same lights. You can see how the colors change but also the amount and type of anemone growth which will throw off alignment software.

Below is a summary of the eleven dives with progress at each step. You can click on the “Blog Post Link” screen capture for each dive to get a summary of that dive. The “Coverage Diagram” graphic shows pictures from the specific dive that were used in the final model. It does NOT show all the pictures taken on a specific dive.

Dive/Date	Coverage Diagram	Blog Post Link	Notes
1 (6 Oct 2022)			Stern area including cargo hold and stern gun.
2 (13 Oct 2022)			Bridge section to add on photos from dive 1.
3 (4 Nov 2022)			Bow section primarily.
4 (15 Nov 2022)			Amazing vis day and got most of the ship but had alignment issues
5 (28 Nov 2022) “The Butterfly Effect”			Worked on the hull section which is tricky due to the lack of features.
6 (17 Dec 2022) “Seaweed Snafu”			Problems begin dealing with the seaweed. Photos align to nowhere.
7 (25 Dec 2022) “Cave Line & Cookies”			Tried using cave line to connect photos. Only used a few photos in the end but critical to align engine room photos from dive 8.
8 (8 Jan 2023) “Seaweed Shenanigans”			Tried using a diver (Lauren) to align photos. Didn’t work. Took more photos of the engine room for the final model.
9 (7 May 2023) “A View From Above”	No photos used in final model	No model created	Tried photos from far above the wreck hoping that software would detect features “outside” the seaweed to align. Didn’t work.
10 (10 Nov 2023) “Secret Mission”			I was joined by friends (you know who you are) to do some “work” and then did a follow-on dive.
11 (3 Dec 2023) “Final Dive”			I had alignment issues from Dive 10 so I needed to use a different method to get alignment.

Next Steps

I have done another dive since this model (dive 12). There are some “blank spots” on the wreck and some areas that are very “green” that I would like to fill in and unify the color. However, the photos from Dive 12 didn’t quite align correctly. At this point, I might do some touch up work in the future but I don’t have any specific plans. The model will have to stay the way it is for a while with the flaws that it has.

I’m also going to encourage Drew Wilson to build a physical model that we can put on the Marissa to help with dive planning and showing the current state of the wreck to divers.

Thanks To…

This was my most challenging photogrammetry project by a long shot — but I also learned a ton. It should not have been as complex as it was but the seaweed and changing conditions were enemies to quicker progress. I want to thank:

1. Lora, Chris, Rod and Captain & Scout at Marissa Charters. All of the dives were done on the Marissa and they always give me the flexibility and time to do the work. They also kept encouraging me to keep going even when I was ready to give up.
2. Tyler for the encouragement to work out the problems to finish the model.
3. The team of divers on Dive 10 (you know who you are) that helped me get over the final obstacle. Thank you so much!
4. Lauren Martin has been a great and understanding dive partner on quite a few of these dives. She helped on Dive 8 when we tried to use a diver “pose” and has been on quite a few of the other dives helping out.
5. My wife most of all for her patience and listening all the time about “this damn Yukon wreck project” that I spend countless hours toiling over on a computer screen.