Make OpenMVS work at jupyter

尝试高清ColMap逻辑

Feature Extractor 特征点提取

• --SiftExtraction.use_gpu 0
  • You cannot use the feature extractor in GPU mode without an attached display
  • issue
• --ImageReader.camera_model SIMPLE_RADIAL
  • camera model
  • Default is SIMPLE_RADIAL, you can choice one of
    • PINHOLE
    • RADIAL
    • OPENCV
    • SIMPLE_RADIAL_FISHEYE
    • RADIAL_FISHEYE
    • OPENCV_FISHEYE
    • FOV
    • THIN_PRISM_FISHEYE

colmap feature_extractor \
   --database_path $DATASET_PATH/database.db \
   --image_path $DATASET_PATH/images\
   --SiftExtraction.use_gpu 0

特征匹配

colmap exhaustive_matcher\
--database_path database.db

Mapper

mkdir sparse
colmap mapper\
--database_path database.db \
--image_path /opt/OpenSfM/data/dinosaur/images \
--output_path ./sparse

Model Coverter

• the out put type could be one of {BIN, TXT, NVM, Bundler, VRML, PLY, R3D, CAM}

colmap model_converter \
--input_path sparse/0 \
--output_path model.nvm \
--output_type NVM

Convert to mvs file

• InterfaceCOLMAP
  • need cameras.txt and so on at sparse with PINHOLD camera mode
  • you can set it by colmap model_converter --output_type

InterfaceCOLMAP \
--input-file $(pwd)/$PROJECT/ \
--image-folder $(pwd)/$PROJECT/undistorted_images \
--output-file $(pwd)/$PROJECT/model_colmap.mvs

• InterfaceCOLMAP

InterfaceCOLMAP \
--input-file $(pwd)/$PROJECT/ \
--image-folder $(pwd)/$PROJECT/undistorted_images \
--output-file $(pwd)/$PROJECT/model_colmap.mvs