I had another dive on the Yukon and had planned on finishing the front of the wreck, starting the bow and starboard side and try to get some of the starboard edge done.
The starboard edge is probably one of the two hardest tasks for this model. The reason is that currents often flow up and over the wreck which makes staying in one place and taking photos that are consistent and equally spaces is challenging.
Dive 2 – Recap & Summary
As of Dive 2, this is where the model stood:
The model consisted of 2,040 pictures.
Dive 3
I made the third dive on Saturday, Oct 30th and spent about 45 minutes on the wreck. My goal was to finish the front end of the ship, start on the bow, and make progress on the starboard edge near the front of the ship.
The dive started out with pretty good visibility. I was the first person in the water and had planned on spending my dive plus the surface interval on the Yukon and skipping the second dive on the reef. There were a couple other divers already in the water from a private boat.
I started shooting near the port edge of the ship near the front of the bridge to make sure I had enough overlap to connect with the other photos. When I got to the bow, I decided to shoot a couple wide angle shots since the visibility was so good. Here are a couple that include the bow, the infamous dolphin cutout in the bow and couple of the guns. (Note: these were shot at f5.6 since I didn’t really care about corner sharpness and was shooting pictures for a model).
At the end of the dive, the water changed temperature and the visibility literally went to 5-10 feet from the previous 30+ feet! It happened in a period of just a couple minutes. I was happy that I had put a strobe at the bottom of the downline. I could have found my way back because I knew where we were tied off, but it gave me a good feeling when I moved toward the stern and saw my blinking strobe in the distance.
Dive 3 – Building the Model
Part of the problem I had (and I expected this) is that as I got close to the starboard edge or around a big structure like the bridge, the currents would push me around and it would be hard to keep a constant distance. This is evidenced by the “hot spots” in the resulting model where I was pushed closer and the video lights illuminated the red colors.
Attempt 1
My first attempt was to add the pictures to the existing model and align them. This actually worked pretty well. Here is the resulting sparse cloud:
It consisted of a little more than 3,000 pictures.
When I went to build the dense cloud, I started to have problems. It would basically only build what existed before. I had checked the normal things that cause this (not resetting the region, etc) and I tried it a couple times but it just wouldn’t work.
Attempt 2
At this point, I decided to take another approach. I would start a new model with all 3,000 photos. I ran the alignment process on “low” accuracy (which took a couple hours) and it didn’t work! It would only align about two-thirds of the photos:
Here is what the resulting sparse cloud looked like:
It was obviously missing the back half of the wreck. I had a theory on why: since the back portion was done on the first dive and there weren’t as many photos, maybe I used medium accuracy alignment? Therefore, I decided to run another model using all photos but with medium alignment.
Attempt 3
I cleared everything out and started the alignment process using medium accuracy. With 3,000 photos, that is quite a task. In fact, it was looking like it would take more than two days!
Okay, that was NOT going to be an option. It was back to the proverbial drawing board.
Attempt 4
In the past, sometimes the alignment process doesn’t result in the photos being aligned in the “first pass.” If you then “reset alignment” on the un-aligned photos and then try aligning them again, it can sometimes work. I decided to take this approach.
I started over and re-ran alignment with low accuracy and it resulted in the same two-thirds of the photos being aligned. Software is at least constant. I then manually reset alignment on chunks of photos at-a-time and then started aligning them. It worked!
I then built the dense cloud, solid model, and texture. The dense cloud and solid model were built on medium quality.
Dive 3 – Final Model
Here are some screenshots of the model after the third dive and a link to the on-line model on Sketchfab.
As noted above, you can see the “hot spots” where I got too close to the wreck due to the currents pushing me into it.
Sketchfab Model
Next Steps
I have a dive scheduled but the weather is getting very unpredictable so I’m not sure the next time I will get to dive the wreck again. My current plan is to finish the rest of the top portion of the bridge (and possibly the top portion of the smokestacks amidships) and then get to work on the starboard side and the starboard edge.
This is going to present a few complexities (current and seaweed moving) so it might be a “bust” but I’m not sure.
I also need to take another look at photo alignment near the starboard edge near the sand at the front of the guns. It looks a bit “off” and like the sand moves up into the ship structure. A project for another day…