Reconstruction parameters

The default parameters are optimal for most datasets. Also, many parameters are exposed for research & development purposes and are not useful for users. A subset of them can be useful for advanced users to improve the quality on specific datasets.

The first thing is to verify the number of reconstructed cameras from your input images. If a significant number are not reconstructed, you should focus on the options of the sparse reconstruction.

Sparse reconstruction

1. FeatureExtraction: Change DescriberPreset from Normal to High If your dataset is not big (<300 images), you can use High preset. It will take more time for the StuctureFromMotion node but it may help to recover more cameras. If you have really few images (like <50 images), you can also try Ultra which may improve or decrease the quality depending on the image content.

2. FeatureMatching: Enable Guided Matching This option enables a second stage in the matching procedure. After matching descriptor (with a global distance ratio test) and first geometric filtering, we retrieve a geometric transformation. The guided-matching use this geometric information to perform the descriptors matching a second time but with a new constraint to limit the search. This geometry-aware approach prevents early rejection and improves the number of matches in particular with repetitive structures. If you really struggle to find matches it could be beneficial to use BRUTE_FORE_L2 matching, but this is not good in most cases as it is very inefficient. (Guided Matching can help in cases, where a single scan is broken into two pieces, one much larger than the other #1801)

3. Enable AKAZE as DescriberTypes in the FeatureExtraction node (it will auto-apply the setting on FeatureMatching and StructureFromMotion when using MR2021 or newer). It may improve especially on some surfaces (like skin for instance). It is also more affine invariant than SIFT and can help to recover connections when you have not enough images in the input.

With Meshroom 2021.0 dsp-sift has been added. It performs better that SIFT and AKAZE in many cases.#136

4. To improve the robustness of the initial image pair selection/initial reconstruction, you can use a SfM with minInputTrackLength set to 3 or 4 to keep only the most robust matches (and improve the ratio inliers/outliers). Then, you can chain another SfM with the standard parameters, so the second one will try again to localize the cameras not found by the first one but with different parameters. This is useful if you have only a few cameras reconstructed within a large dataset.

Dense reconstruction

1. DepthMap
   You can adjust the Downscale parameter to drive precision/computation time. If the resolution of your images is not too high, you can set it to 1 to increase precision, but be careful, the calculation will be ~4x longer. On the contrary, setting it to a higher value will decrease precision but boost computation.
   Reduce the number of neighbour cameras (SGM: Nb Neighbour Cameras, Refine: Nb Neighbour Cameras) will directly reduce the computation time linearly, so if you change from 10 to 5 you will get a 2x speedup. A minimum value of 3 is necessary, 4 already gives decent results in many cases if the density of your acquisition process regular enough. The default value is necessary in a large scale environment where it is difficult to have 4 images that cover the same area.

2. DepthMapFilter
   If you input images are not dense enough or too blurry and you have too many holes in your output. It may be useful to relax the Min Consistent Cameras and Min Consistent Cameras Bad Similarity to 2 and 3 respectively.

3. Meshing
   If you have less than 16G of RAM, you will need to reduce the Max Points to fit your RAM limits. You may also augment it, to recover a more dense/precise mesh.

(#195)

If you notice holes around an object in 2021.0 that were not there in v2020.1.1, it may come from the new post-processing of the tetrahedralization. To get previous behavior, change the following parameters on the Meshing node (in the advanced params) (*):

nbSolidAngleFilteringIterations=0

Tetrahedron Neighbors Coherency Nb Iterations=0

4. MeshFiltering
Filter Large Triangles Factor can be adjusted to avoid holes or on the other side to limit the number of large triangles. Keep Only The Largest Mesh: Disable this option if you want to retrieve unconnected fragments that may be useful. (Disabled by default in v2021)

5. Texturing
   You can change the Texture Downscale to 1 to improve the texture resolution.