210202 Koji Yonekura, RIKEN SPring-8 / Tohoku University
Derived from detect.py in yolov5
cnvxtalpos2Nav.py: Derived from convLM2DIFF.py (by Kiyofumi Takaba)
210403 Version 1.0
- Download yoneoLocr-main.zip from https://github.com/YonekuraLab/yoneoLocr.
- Extract the zip file and put the whole directory as yoneoLocr in C:\ProgramData\ of a camera control Windows PC.
- Set the property of batch files to “full control” from the Security tab if needed.
- Install CUDA Toolkit 10.1 and cuDNN 10.1 for a K3 control PC if the operating system of the PC is Windows Server 2012R2. Newer versions of CUDA and cuDNN are available for Windows 10.
- Install Microsoft Build Tools for Visual Studio (vs_buildtools) if needed.
- Install ImageMagick.
- Launch Anaconda Prompt. Make and activate an Anaconda environment as,
> conda create -n yolov5-4.0 python=3.8
> conda activate yolov5-4.0
- Go to the yoneoLocr directory and install python libraries as,
> conda install pytorch==1.7.1 torchvision==0.8.2 cudatoolkit=10.1 -c pytorch
> pip install -r requirements.txt
- Put shortcuts, yoneoHole, yoneoXtal, yoneoDiff, and yoneoLowmagXtal on the desktop.
- Launch yoneoLocrWatch.py from the shortcuts.
- Select running mode.
--object hole / xtal / diff / lowmagxtal
- A confidence threshold for object selection in hole and lowmagxtal modes. Default 0.4.
--conf-sel 0.4
- Delete output file showing objects enclosed with boxes. Default: no.
--delout yes / no
- Include ice crystals for positioning in xtal mode. Default: no.
--ice yes / no
- Other options in the original script detect.py in YOLOv5 are also available.
- A weight file is included only for "hole" due to limitation of the file size at the github site. Other weights are available from our web site.