Releases: gajdosech2/scan-filtering
Thesis Release
1.0 f1 in eval
Beta Release
0.9 f1 in eval
HIROVisualizer
v0.4.2 cogsconverter -1/0 fix, zmap
ŠVK Release
Usage
- copy all the .cogs files you wish to filter into the process subdirectory
- run FILTER.bat
- filtered files are in the newly created result folder
COGS Converter
COGS Converter is an utility (console app) meant for converting scans (organized pointcloud) from .cogs format to other, more general and simpler formats, especially for further processing in (machine learning) pipelines.
Supported features in this version, with examples:
- convert .cogs to set of images (normal map, depth map, intensity map) (NOTE: all values are scaled to 0-1 range, inform me if you for some reason need an optional parameter to disable this auto-normalizing):
CC.exe --generate INPUT_FILE_PATH EXPORT_PATH(optional)
ex.:
CC.exe --generate scans/zoo_000.cogs images/
- additionaly to set of images, generate a ground-truth mask for filtration, or some other binary segmentation problem, if you have the ground-truth .cogs:
CC.exe --generate INPUT_FILE_PATH EXPORT_PATH GROUND_TRUTH_PATH
ex.:
CC.exe --generate scans/zoo_000.cogs "" scans/truth_zoo_000.cogs
- convert .cogs to .xyz file, which is just a text-file with coordinates of each point in separate rows. Optionally, you can generate an .outliers file from ground-truth scan, which is just text-file of zeros and ones. This format is used for example by https://arxiv.org/abs/1901.01060 and you can visualize it using http://lidarview.com/ :
CC.exe --xyz INPUT_FILE_PATH EXPORT_PATH(optional) GROUND_TRUTH_PATH(optional)
ex.:
CC.exe --xyz scans/zoo_000.cogs "" scans/truth_zoo_000.cogs
NOTE 1: Scans are always trimmed in the process (unnecessary padding of zeros from all sides is removed). This can be easily turned-off in the code, message me if you need assistance and need a binary with this feature disabled by parameter.
NOTE 2: The source code should be build-able on Linux system, make the binary yourself or inform me, if you have any problems.
NOTE 3: If you need to convert .cogs file to some other format, you should be able to implement your own converter by inheriting from the base abstract ScanFormatter class. Message me if you need assistance and feel free to contribute your converter.
04.10.2020
INSTRUCTIONS
System prerequisites
Windows with NVIDIA:
- NVIDIA GPU Driver 418.x or higher
- CUDA Toolkit 10.1 https://developer.nvidia.com/cuda-toolkit-archive
- cuDNN SDK 7.6 https://developer.nvidia.com/cudnn,
- https://towardsdatascience.com/installing-tensorflow-with-cuda-cudnn-and-gpu-support-on-windows-10-60693e46e781
- https://www.tensorflow.org/install/gpu
Ubuntu with NVIDIA:
- not available at the moment
Ubuntu with AMD RoCM:
- not available at the moment
Python Environment
- Python 3.7 or higher
- python libraries:
pip install tensorflow
pip install keras
pip install numpy
pip install matplotlib
pip install imageio
Usage
- copy all the .cogs files you wish to filter into the process subdirectory
- run FILTER.bat
- filtered files are in the newly created result folder