Plus additional goodies, such as
- equirectangular camera support
- OpenGL 3.1 support
- more accurate voxel pruning for training
- cleaner GUI

When building with MSVC, generator expressions are required to utilize the correct DLSS library files -- else the build fails

- SIMPLE_RADIAL_FISHEYE
- RADIAL_FISHEYE
- OPENCV_FISHEYE

Add the path to a vocabulary tree when chosen "vocab_tree" as colmap_matcher
using vocabulary tree give better results https://demuc.de/colmap/

handle reflection case with small noise

handle exception for the opposite direction input

- orientable crop volume
- extra python bindings
- better visualization of cameras
- a number of fixed crashes

When passing parameters to Colmap2Nerf the system calls example: --video_in "D:/DataSet Videos/Video.mp4" Won't recognize path as a whole paramter due to Absence of Double Quoations, so argparse interpret it as  "D:/DataSet" only and "Videos/Video.mp4" as an another argument.

Add 'SIMPLE_PINHOLE' and 'PINHOLE' camera models.

add a parameter to take start and end time, to extract images from the
video only within the given time slice

Signed-off-by: Ameya Shenoy <shenoy.ameya@gmail.com>

Correcting previous mangling of the indentation

From numpy 1.7, the type of implicit casting from L:271 `totp += p*w` produces an error: 

```
TypeError: Cannot cast ufunc add output from dtype('float64') to dtype('uint8') with casting rule 'same_kind'```

To avoid `totp` and `totw` should be initialized as `dtype('float64')`.

The "split" function, which was supposed to determine the filename, fails, if spaces are contained in the filename itself.
I fixed it by splitting the "line" variable by everything BUT the filename.

fix regression in colmap2nerf.py that caused it to fail to find the center of the cameras point of interest, due to a sign-flip in the closest-point-of-approach function. this should make it more robust at correctly centering new datasets