Convert Camera Parameters [model of colmap]

Try Colmap with cuda(gpu)

• enable gpu with args --SiftExtraction.use_gpu 1
  • but throw qt error, need x display
• close QT gui by export QT_QPA_PLATFORM=offscreen

  • but throw these error below

    *** Aborted at 1628840833 (unix time) try "date -d @1628840833" if you are using GNU date ***
    PC: @     0x7f07a36bbbf8 (unknown)
    *** SIGSEGV (@0xe0) received by PID 19386 (TID 0x7f07b1686900) from PID 224; stack trace: ***
        @     0x7f07b0919980 (unknown)
        @     0x7f07a36bbbf8 (unknown)
        @     0x7f07a36bbff8 (unknown)
        @     0x7f07ae65407f QOpenGLContext::create()
        @     0x5601b94e4b93 (unknown)
        @     0x5601b9391777 (unknown)
        @     0x5601b9396eec (unknown)
        @     0x5601b925b4a7 (unknown)
        @     0x5601b924c863 (unknown)
        @     0x7f07ac9edbf7 __libc_start_main
        @     0x5601b925073a (unknown)
    

• wait for issue#1273 response

Convert camera parameters from opencv-matrix file to other type

Jupyter Code

Attation

• This need pip install opencv-python but conda install
• For now, the version is opencv-python==4.5.3.56

Code

fs = cv2.FileStorage("camera.xml", cv2.FILE_STORAGE_READ)
root = fs.root()
for key in root.keys():
    intrinsic = root.getNode(key).getNode("camIntrinsics").mat()
    external = root.getNode(key).getNode("camExternals").mat()
    width = root.getNode(key).getNode("imageSize").getNode("Width").real()
    height = root.getNode(key).getNode("imageSize").getNode("Width").real()

• convert for 空三
• jupyter lab

Figure out the camera model of colmap

Camera model list at here

• SIMPLE_PINHOLE, PINHOLE
  • The easiest model, means 相机图片没有任何畸变，主要指内参
• SIMPLE_RADIAL, RADIAL
  • 默认选择，这个会自动估算图片畸变并计算相机内参
• OPENCV, FULL_OPENCV
  • 引申搜索到了google nerfies
  • 一个3D拍照应用
  • 
  • 
  • prepare training data - colab
  • training data - colab
• SIMPLE_RADIAL_FISHEYE, RADIAL_FISHEYE, OPENCV_FISHEYE, FOV, THIN_PRISM_FISHEYE
  • 用于鱼眼镜头