LoMa Nodes

Meshroom nodes using LoMa (rez package lomatch) as a replacement for FeatureExtraction + FeatureMatching.

Meshroom Nodes

LomaFeatureExtraction

Detect and describe keypoints for each view of the input sfmData. Writes <viewId>.roma.feat/.desc (AliceVision features, roma describer type) and <viewId>.loma.npz (LoMa keypoints/descriptors used for matching).

LomaFeatureMatching

Match the extracted features on a list of image pairs with the LoMa matcher. Writes AliceVision matches (roma describer type).

Typical graph

CameraInit → LomaFeatureExtraction → ImageMatching → LomaFeatureMatching → StructureFromMotion

• ImageMatching: featuresFolders = LomaFeatureExtraction output, method Exhaustive or Sequential (vocabulary tree requires SIFT).
• StructureFromMotion: featuresFolders = LomaFeatureExtraction output, matchesFolders = LomaFeatureMatching output, describerTypes = roma.

Weights

LoMa, DaD, DeDoDe and DINOv2 weights are downloaded by torch.hub on first run (~/.cache/torch/hub, or $TORCH_HOME/hub).