At its core, photogrammetry sounds like a simple process. Take a bunch of pictures covering every angle of your subject. Drop these photos into a software system and wait for the results, As a concept this is basically true. However, by conducting objective work reviews, determining better techniques and experimentation can often lead to better results.
It has been over a year now, learning photogrammetry and all the associated techniques in order to bring something from the real world into a virtual/game world. The initial results were very interesting and the time and effort per-model was comparable to traditional modeling methods. However, I tend to be a perfectionist. If I see something that I deem an error, then I focus on that problem until I can correct the issue. Not a bad way to work but, it can often times slow down the process. I attempt to view my final product from an objective point of view. This blog explains one aspect that clearly shows the benefits of conducting your workflow using an objective point of view.
This site contains a large number of images. You may be surprised to learn that none of the images are photos. Each and every image is either a 3d render or shot within the UnReal 4 Engine or alternate game engine. Every image on this site has at least one photogrammetry based 3D model.
Over the last few months, I have determined that how you take your photographs is just as, if not more, important than lighting or anything else. I have figured out consistent techniques that remove/limit texture 'smearing' as well as other 'base mesh' issues. The image above shows the camera positions used to capture a common rock using my initial methods. The rock was photographed from three to four feet away and at a downward angle. The model looks good. however, this image shows just how much 'smear' occurs in the texture;
You will note the camera distance from the subject in the first photo. I determined that even at 20.1 megapixels and proper settings, my camera and my technique caused the texture smear issue. After making adjustments and a bit of practice I achieved this result for with a new model;
You will probably not notice an initial huge change. However, I shot this object at a much closer range. Only moving 10 to 12 inches from the subject. The model looks good and the mesh is very clean. The most interesting result from this experiment is the texture;
You will note that the texture now looks much sharper and there is far less 'smearing' evident in the source images. The mesh is still solid and now the texture appears much more detailed and realistic. Even the color information is closer to the rock's proper Physically Based Rendering (PBR) settings, reflective and roughness/specular ranges.
I output each of these textures at an 8k resolution. By using this more detailed approach, I have eliminated the majority of smearing and my final result, (rendered at a 2k resolution) contains far more detail and data then I was previously able to achieve with the final game-ready texture maps even using a texture cage method of mapping.
In conclusion, adjusting my technique lead me to more accurate and better visual results. This may be obvious to many of you that follow my blogs and information but, this simple act of objectively reviewing your work, identifying issues and working out a plan to correct those issues must be done. Remember, there is no such thing as 'perfect' but, we can at least strive for it.