A gallery of failures! and a write-up on this term

I am essentially confirming to have reached a dead end with my separated mesh idea! So I used the auto mesh decimation tools and made a singular mesh version as well.

Both reconstructions of the data are, well, bad.

Here is the separated mesh in Unity


And here is the single decimated mesh in Unity

In no particular order, here are some fun screenshots of various issues

So what exactly happened? The problem is that I've treated this shooting this small room as single object. My original thinking was that having a 'blank canvas' of a room as a single reconstruction would be a good idea. Turns out, it was not. It proved to be difficult to separate the mesh after the fact, and instead, I should have planned on collecting separate data from the beginning.

Here were all my issues that had to do with the complexity of the subject:

• Reprojection didn't work with manual retopo (with mesh seperations) and manual UVing. 
• inaccurate reprojection
• UV unwrapping didn't work 
• due to tri based mesh
• due to manual retopo not working
• Map baking such as delighting didn't work
• due to UVs being messy

I don't think I have much more to gain from this particular project without either reshooting the data, or preferably, scouting a new location with better planning. 

Particularly after reading up on the Unity Documentation from the last few posts, I feel that I will have much more to learn from a project with better planning. I think my issue these past few weeks was getting stuck on this particular data set.

However, I have had a lot of valuable takeaways from this term.

1. Research in Software
• 3DF Zephyr has been great to work with. The major painpoint is that it is missing UV unwrap tools.
• I have settled into Unity being my engine of choice
• Substance Painter is my friend for all things manual, but I am leaning towards using xNormal and Unity Delighting tool for automated map generation.
• Need to continue in researching Substance Designer, which also has all the needed map bakers AND is used for procedural materials (lots of documentation on procedural photogrammetry textures).
2. Research in Data Acquisition
• I learned that DSLR cameras are more accurate, versatile, and accessible than the available scanner hardware.
• I learned there are tons of directions to research in better shooting methods. For example, building new camera rigs, or just documenting best practices.
• I learned that this is the fun part for me, so much that I have continued in a side project of collecting 3D portraits of anyone who will sit for me. (Should really post these somewhere...)
3. Research in Data Processing
• Yes, pre-processing is a thing, but considering how fast Zephyr is, I've found that often its faster to just reshoot with adjusted settings.
• ZBrush Dynamesh is the best remeshing tool, ever. Configure edgeflow and resolution, and output quads.
• Another painpoint is Zbrush has a problem with exporting bad meshes that Zephyr can't read. 
• So far post-processing has been the worst part of the pipeline so far. I find myself jumping between 3-4 software packages just to use one of their tools. It will be great in the future to have a end-to-end solution for photogrammetry game assets.
4. Furthering the field
• There is SO much that can be published at this point. Any project utilizing this kind of technology can contribute by documenting best practices, new capture techniques, new processing / geo reconstruction techniques. In fact, these techniques might not be 'new' but they may borrow from other fields and be applied instead to games/interactive production.
• With that being said, there are also SO many existing articles, both scholarly and non scholarly. I am kicking myself a little for not spending more time following the existing documentations... it's hard to balance R+D time for a project. So easy to get stuck reading forever, and so easy not to do enough to begin with.