User Guide for the GUI Client

Steps:

• Set a directory where results will be saved with menu File -> Set Project Directory. Note that previous results in this folder may be overwritten. So back up them or use a different folder if you want to keep previous results.

• Add your images with File -> Add Images. You can select multiple images in a batch. There will be a dialog for optical parameters. If the added images are shot with the same camera and locked focus, you should tick Tied Optics, otherwise uncheck it. The focal length (fx, fy), and optical center (cx and cy) are in pixels and the client will try to infer the values from EXIF info. The optical center coordinates start from topleft corner (rather than center) of the (0,0) pixel. So a ideal pin hole camera will have (cx = width * 0.5, cy = height * 0.5). The distortion parameters are defined in OpenCV convention and defaults to zero. If you have pre-calibrated parameters, that will help improve precision. For images shot with some DJI drones, e.g. DJI Phantom 4 RTK, the pre-calibrated parameters in the XMP metadata will be used.

• Optionally set sparse reconstruction (image alignment) settings with Project -> SFM Settings.

  • Optical Model:
    • F1 means fx = fy constraint during optimization.
    • F2 means optimize fx and fy without the equality constraint.
    • C2 means optimize cx and cy.
    • D2 means optimize k1 and k2.
    • D5 means optimize k1, k2, p1, p2 and k3.
  • Fixed Optics: check if you are confident about the pre-calibrated optical parameters you provided.
  • Shutter Model:
    • Global: global shutter. Also works for rolling shutter if camera motion/rotation velocity is low while shooting.
    • Rolling1DLoc: rolling shutter with one degree of freedom. Rotation is set to zero and the camera moving direction is infered from gimbal pitch, with assumption that the drone moves horizontally. “Loc” here means duration optimization, the moving direction is fixed relative to camera local coordinate system.
    • Other rolling shutter modes are available but not yet exposed in the client.
  • Key Points: the number of key points used in most of the steps. Note that we try to extract much more than what is set here (~4 times) but will only pick this amount for most later steps.
  • SIFT Upsampling: upsample the image for the first octave. The original SIFT paper has this. We default to no.
  • Use GPS: use GPS information in the image XMP and EXIF. Default to yes when GPS in XMP is detected and no when only EXIF GPS info is available, or no GPS at all. Note that GPS in EXIF information has low resolution (typically 1 angular second, or 1/100 angular degree).

• Optionally save the config with File -> Save SFM Config, so in case of failure you can reload the above into with File -> Import SFM Config.

• Start sparse reconstruction with Project -> Align Images. On finishing, a sparse 3D model will be shown. If multiple models are generated and cannot be merged, the largest one will be shown. The sparse model will be saved in the project folder in three formats:

  • RSM - the format used by RapidSFM. Can be opened with rsfm_client with File -> Open Sparse Model.
  • NVM - the format defined by VisualSFM and used by some other software.
  • PLY - the more commonly used format. You can view it with rsfm_client via File -> Open PLY, or use MeshLab.

• Start dense reconstruction for the shown sparse model, with Project -> Densify Point Cloud. On finishing, the dense point cloud will be shown and saved as *_dense.ply.