Create your model
Create the model in Metashape
Great! You now have a bunch of overlapping photos of a sample with varying degrees of quality. If you're like me, you tried to follow best practices, but had to cut corners here and there due to current circumstance. Let's try and make the best possible model with what we have!
This part of the process is pretty straightforward, I would just follow a typical Metashape workflow of importing photos, aligning photos, creating a mesh and building the texture.
Import photos and align photos
Select all photos you need and align them. You might be better off to split the photos into two chunks if you had to flip over the sample to photograph the underside and have a lot of background visible in the photos. However I would try to align photos without splitting in chunks first, as this step is usually quite rapid and can be re-run after you split the chunks. Make sure you inspect the photo alignment closely for any strange alignments
Poor result of photo alignment, where photos grouped together in bunches! In this case, the sample was too small and had too much background visible. The solution was to mask the model, more about that in the next section!
Fixing poor alignment
In order to fix poor alignment, you can look out for tie points that look like they're in the wrong place, giving a strange outline to the object. You can see that in the video above. Here is a quick checklist:
- Identify the tie points that look wrong
- Select the offending tie points with the lasso tool
- Right click on tie points - 'filter photos by tie points'
- Select the photo and right click on photo - 'Reset alignment'
- Align photos one by one by right clicking on the photo - 'Align photo'
Build mesh
Build the mesh, this will generate the surface of the model. For the source data, you might find that the sparse cloud is good enough as a mesh and can be generated nearly instantaneously. Using depth maps will build a much more detailed mesh. Your choice really depends on what effect you are going for. A highly detailed mesh might look better if you use active lighting for rendering, although this is subjective. You might need to use depth maps to identify small or intricate details. The trade-off is that a mesh with less detailed mesh (a lower polygon count) will be quicker to load and to render - however I've found that this is rarely an issue these days.
Build texture
Building the texture will generate the image that will be draped over the model. This is the step where you find out what detail you will get in your final model. You might want to duplicate the model and or/chunks in order to generate several textures using different settings and different masks in order to compare them. Some rules of thumb:
- I've found that the using the Spherical mapping mode results in a nicer post processing effect on your models, such a lighting.
- Your texture size (which corresponds to the size of the image that will be draped) should be 4096 or less. Some phones or lower end devices won't be able to run model with an image that is greater than 4096 px. However- if you use sketchfab, textures get downsampled when your model is uploaded so it might be ok to go higher.
- different blending modes will produce different effects, and it's worth trying more than one. I've found that mostly that Mosaic tends to produce the effect that I want, but sometimes I've preferred the result when using Average.
Example of texture Image
A 4096px texture image. This image is tied to the model using a map of coordinates, called U,V coordinates .
You made it this far, meaning that you've not got a model in a some state. If you're happy with the result, you could jump straight to the final steps. However, there is a chance that you see some parts of your model that can be improved. In that case, try the next section: