YoloCR, optimized for Docker.
YoloCR is a convenient OCR script, forked from https://git.clapity.eu/Id/YoloCR
This is a docker-ized version of that project, trying to make as few changes as possible to the original files.
Why ? Well, considering the many dependencies to build and prepare to set this up (see original README, or the original installation script), I just did not want to pollute my WSL installation or create a heavy dedicated environment/VM just for this, especially considering how low on space I was.
So, this Docker project achieves to make it fit in a ~200MB compressed image, instead of a ~9GB VM. I am quite happy with that, to be honest. I used Docker multi-stage build to achieve such a result, by building the sources separately and then installing the built packages on the final image, without keeping any build dependencies or leftovers. You can see that in the Dockerfile.
Most of the usage of the project was automated, instead of having multiple interactive steps, to match unattended scenarios with a container deployed on demand. It works great on Azure, to automatically process files added on a drive for example !
I wondered whether to build FFMPEG from source to size it down even more, but ultimately decided against it. You can do so if you have very specific needs and know your input format in advance, it can save you a lot of space !
Preparation
You will have to read the original README, to learn how the original project works first.
You can however skip the installation part, and you will notice that you need to first determine the required parameters for your video : thresholds, pixel areas, and some others. Note that the main author is french, and they made the parameter names as well.
Without having a GUI in the container, tools like this can help you find the relevant parameters : both for colors thresholds or crop box coordinates.
With the parameters figured out, you can create a .conf file to store these, following this template file, or just write them down somewhere else to then provide them as environment variables while running the tool.
Finally, Prepare a data volume that will act as the input/output directory. Put the video file(s) you want to OCR in there, and an eventual configuration file to use for these, and the container will write OCR results in *.srt format in this volume.
I also assume you are a bit familiar with Docker and Linux. If you are using this on Windows, you are advised to do so through WSL, for example if you want to mount a folder as a volume in your docker container.
Usage
Start a container with your data volume mounted to /data, and with the parameters you want as environment variables or in a config file in the data volume. In this data volume, all .mp4 files will be treated as input files, and any .conf file will be imported as a configuration for all the input files in the volume.
If you ever need a reminder of the standard YoloCR parameters, you can refer to the template configuration file, which holds both the default values and comments about how to use these parameters.
docker run -it -v:"THE/FOLDER/YOU/WANT/TO/PROCESS":/data amine1u1/yolocr
or
docker run -it -v:"THE/FOLDER/YOU/WANT/TO/PROCESS":/data amine1u1/yolocr -e "DimensionCropBox=1344,150" ...
Once done, the container will create a folder with a .srt subtitle file for each input file, with the OCR results according to the parameters you set.
Two additional environment variables are available :
FILEEXT, which defaults to "mp4", is the comma separated list of fileextensions to treat as input files.KEEPDATAis an optional environment variable. If set to any value, the temporary working data, such as time codes, ScenesChanges, and working screenshots, will be copied to your data volume on completion. Helps for troubleshooting.
Credits
Thanks to Yuri Zero for his original work on YoloCR , which is embedded in this hard fork, keeping his original commit history