This project uses JDK 17.
This project aim to recognize conversation text area in PJSK event story record videos and
automatically generate .ass files for the video to add subtitles.
This is a rewritten project based on Jeunette / ASS-automation-java, this project improves runtime, stability, and readability of the previous project, so that further improvements and modifications can made easier.
This project currently includes opencv_java460.dll and opencv_videoio_ffmpeg460_64.dll library files in release zip files. The program is written, compiled,
and tested on windows 10 x64, other platforms and architectures would require different library files.
You will need a .mp4 video file, recording one PJSK event story, and a .json file of that event story from the game assets. The .json file can be found at
pjsek.ai thanks to Erik Chan.
If you selected both files, click the run button on the gui to run the program. The gui will present you with the current step of the task and the total tasks. After
the task completes, a .ass file will be created in the same directory of the video file with all parsed data. In case of failure during task, an error message should be
presented on the gui.
This file defines user settings, including
sampleAssFilePath: defines the path to the .ass file including headers, styles, and predefined formats, see the sample.ass file in the repository.
resolutionMaskConfigFilePath: defines the path to all the resolution mask files, see below.
resolutionMaskConfigFileNames: defines all the file names of the resolution mask files. Each resolution mask file defines masks and styles used for text mask and transition
text mask under a certain resolution.
This file defines character styles, the format is "characterName" = "characterStyleName", the names should be wrapped by double quotes so that nothing is escaped.
There is a default setting, all unknown characters in the story will use the default style.
You can run the .jar file in console, there should be some logging information printed. You can also start the .jar file with the arg --debug, the loggers will use
properties defined in logging.properties and generate a log file, that log file can be used to debug the program.
The keyframe detector will try to parse the video frames in sequential order. If a conversation text is detected, it will try to skip ahead by an estimated amount of frames. This makes it generally faster and require less CPU and RAM. However, note that this method is not robust, as if one detection fails, all following detection will also fail, but I have tried my best to make the detections accurate enough.
The multithreading detector is not implemented yet and requires more time. Currently it is disable by the code.