When I attempted to use COLMAP in ODM instead of OpenSfM, I performed undistortion on the reconstructed data and exported the undistorted reconstruction files to the NVM format. However, I noticed that the undistortion results differed from those of OpenSfM, and the final digital orthophoto map (DOM) was incorrect. Upon investigation, I realized that COLMAP’s undistorted camera model includes a principal point, whereas OpenSfM assumes the principal point to be the image center.
To address this, I modified the undistortion code in COLMAP to output images without the (cx, cy) offset, as shown below:
// Copy principal point parameters.
undistorted_camera.SetPrincipalPointX(camera.Width() / 2.0);
undistorted_camera.SetPrincipalPointY(camera.Height() / 2.0);
While the resulting DOM appeared visually correct, its precision was not as high as OpenSfM’s pipeline. Are there any additional steps I should consider when integrating COLMAP into ODM?
The commands I used are as follows:
colmap feature_extractor \
--database_path $DATASET_PATH/database.db \
--image_path $DATASET_PATH/images
colmap spatial_matcher --database_path $DATASET_PATH/database.db \
--SpatialMatching.max_distance 150
colmap mapper \
--database_path $DATASET_PATH/database.db \
--image_path $DATASET_PATH/images \
--output_path $DATASET_PATH/sparse
colmap image_undistorter --image_path IMAGE_PATH \
--input_path $DATASET_PATH/sparse \
--output_path $DATASET_PATH/dense \
--min_scale 1.0 --max_scale 1.0 --output_type COLMAP
colmap model_converter --input_path $DATASET_PATH/dense \
--output_path $DATASET_PATH/dense/scene.nvm \
--output_type NVM
Note: I am using a forked version of COLMAP 3.8 with GPS constraints.