initial commit

README.md

@@ -0,0 +1,22 @@
+# Touhou MMD Subtitle Generator v0.3
+Created by [PatchyVideo](thvideo.tv)\
+Based on [CRAFT](https://arxiv.org/pdf/1904.01941.pdf) scene text detector\
+Translation is done using [Google Translate](https://translate.google.com/) with python package [googletrans](https://github.com/ssut/py-googletrans)
+# How to use
+You need an NVIDIA GPU no older than GTX 700 series
+1. Download `textdet_300k.pth` and `ocr_640k.pth` from release and put them into the project folder
+1. Install NVIDIA Driver
+1. Install CUDA 10.2
+1. Install Python3 environment
+1. Install PyTorch 1.5.1 from [offical website](https://pytorch.org/)
+1. Run `pip3 install -r requirements.txt`
+1. Run `python main.py --out NUL --verbose --skip_frame 1 --video /path/to/video/file` and have fun
+# How to use in real life
+You can specifiy an output srt subtitle file of recognized texts using `--out /path/to/out/srt` parameter\
+No windows will be shown if `--verbose` parameter is not given \
+`--skip_frame 1` controls how many frames to skip in-between two frames used for recognition, higher the value faster the program will run, but the resulting srt file will not be as accurate as setting this value to 0
+# Future works
+1. The current text detector is not very accurate, new model is being trained as you are reading this
+1. Support Korean
+1. Build multithreaded C++ version to fully utilized both CPU and GPU
+1. Integrate into [PatchyVideo](thvideo.tv) so people can use it online to enjoy Touhou videos made by people speaking another language