We had only one model that was within the range of success, made with the same process we have been trying. The only difference was that our subject sat down for the pictures. Our other successful model was also sitting, so its possible that the position is the key, rather than more detailed shots.

Our guess is that the motion of the standing subjects could be the issues. If the subject is swaying even a little bit (possibly trying to keep still, or subconsciously leaning to face the cameraman), it might be enough to throw off the perspective of the software. This could explain not just why the head is missing, but why it appears to be blown up against the walls. We’ll test this theory by trying another take of a person sitting down, and see if the model is higher quality or if we just got lucky.

Another factor that could have gone into this is the better hardware we now have access to. We’ve started using an IDeATe Alienware laptop to do our photo processing, and it’s made the whole process much cleaner. Because of how much stronger the laptop is than the machine’s we’ve been using, we’re able to create higher poly models without sacrificing days to the process.

Now that we have our first successful model, our goal is to use this process to create a full model. We’re going to test what happens if we do a lot of close-up shots on a person’s key features, like their face, and then follow those with full body shots. We hope that getting high-detail captures of key areas will let the software use those areas to better align the other photos it has, giving us the accuracy we were missing in our original tries.

We’ll be back next week with some new tests of this process, and hopefully with a full body model!

This week, we’ve started having our first successes. We redid a partially successful take at higher poly count / texture quality, and got the above model. The proportions are definitely still off, but it’s the first take where you can definitely tell that a human was the subject.

We suspected that the software was having some issues doing the humans as a whole, so we are in the process of trying to create a model from a set of photos focusing on a face. We don’t have a finished model yet, but the point clouds that were generated look very promising. You can pretty clearly see the outline of the head!

During this retake, we also ran into some issues with our software. Both of our licenses for Metashape ran out, so we’ve been trying to find replacement software. One contender that we’ve been trying is Meshroom, which you can see in the shot above. Meshroom is completely free (open source), and produces pretty good results for other people, so we decided to give it a try. Unfortunately, it turns out that you need a Cuda-enabled graphics card for it, or else you get results that look like this:

That’s it for this week! Hopefully our next post will contain some good shots of face models!