Redrobes blog about playing with the technical side of graphics.

In my last entry I mentioned that Meshroom had been released. I had to borrow a friends machine with an nVidia card on it that would handle CUDA as the program will not run without it. The computations required to generate the 3D mesh from the photos are quite intense and with many photos it takes a lot of time. CUDA is the graphics card support which allows the Graphics Processing Unit (GPU) to handle that complex graphics based math instead of the normal CPU. Since GPUs are custom built for this kind of thing then it can do it more easily and faster.

I went to a stately home which had a statue in it and seeing the opportunity, I took a lot of photos of it from all angles that I could. That would be everywhere except for the very top of the head. I made sure to cover all the crevices and undercuts which are normally a hard problem for 3D reconstruction.

Below are three images. The first is one of the 60 photos that I took of the statue. Once Meshroom had processed all of them then it produced a very good 3D model which is shown in the second photo. The third is a GIF of the 3D model spinning.

I think you would agree that this photogrammetry has come a long way from when I started this blog about it. The model is still not ridiculously dense with points but its definitely good enough and since it can map back on the texture from the photos to the model then it looks very good indeed.

I would now like to try this out with some terrain photos and maybe some drone footage but I still do not have a GPU that handles CUDA so it may have to wait for another day. It's on my tick list tho don't you worry !