Write OpenSfM parameters presets for different types of data.

paulinus / OpenDroneMap

OpenDroneMap is a tool to postprocess small Unmanned Aerial Vehicle (sUAS), balloon, kite, and street view data to geographic data. With the current update, we are adding the ability to create orthophotos from drone, balloon, and kite imagery which has GPS ephemeris. Future versions will permit the use of GCPs with no onboard GPS. forked from qwesda/BundlerTools ( https://github.com/qwesda/BundlerTools )

GNU General Public License v3.0

11 stars 5 forks source link

Write OpenSfM parameters presets for different types of data. #9

Open paulinus opened 6 years ago

paulinus commented 6 years ago

basic
large drone datasets (gps based matching, local bundle + global)
linear datasets (order based matching, local bundle)
high accuracy (more features, stricter thresholds, larger images and depthmaps)

dakotabenjamin commented 6 years ago

Currently we could do just local bundle or just global, but not the hybrid type that you mentioned in the call, right?

paulinus commented 6 years ago

We can do hybrid also. For example, run local bundle for every image added, then a global bundle from time to time. For example, every 100 images and everytime the number of points increases by 20% since last full bundle.

bundle_interval: 100            # Bundle after adding 'bundle_interval' cameras
bundle_new_points_ratio: 1.2    # Bundle when (new points) / (bundled points) > bundle_new_points_ratio
local_bundle_radius: 1          # Max image graph distance for images to be included in local bundle adjustment

dakotabenjamin commented 6 years ago

Oh that is perfect.

dakotabenjamin commented 6 years ago

FWIW I tested the hybrid bundle adjustment method on >5000 image dataset and that part worked. It crashes on merging depthmaps though, as I expected.

EDIT: I added 120GB more RAM and it worked fine: