MMD字幕识别v0.5.2版本/MMD-OCR v0.5.2

中文

下载说明

你需要下载MMD-OCR-v0.5.2.7z

硬件要求

你需要至少一张9系及以上的NVIDIA显卡（只要是Maxwell及以上都可以，包括部分9系显卡）
程序会占用2G内存和1G显存。
CUDA计算能力版本：5.2;6.1;7.5;8.2

使用说明

请先更新显卡驱动到最新版
把视频文件拖到mmdocr.exe上，一个对应的srt文件就会在视频的同一文件夹下生成

已知问题

1. 部分字符不在字典内，无法识别。
2. 滚动字幕、旋转字幕和特效字幕处理的很差，不建议用在舞蹈类或PV类视频上。
3. 部分文字水印会被持续识别，污染生成的字幕文件。
4. （重要）不支持竖版文字。
5. 会自动覆盖已有的srt文件，注意备份。
6. 优化不到位，无法利用全部显卡资源。

文件说明

mmdocr.exe：识别程序，不会展示中间识别过程，实际生产请使用该程序。
代码在src文件夹里。

English

Download

You only need MMD-OCR-v0.5.2.7z

System requirements

NVIDIA GPU at least GTX 900 series is required.(Maxwell or later)
Program will use 2G of RAM and 1G of VRAM.
CUDA computability required: 5.2;6.1;7.5;8.2

How to use

Please update your graphics driver to latest version before proceeding.
Drag any video file to mmdocr.exe and a .srt file will be generated at the same folder as video file.

Known issues

1. Some characters are not recognized because they're not in my dictionary.
2. Scrolling texts and rotated texts are handled poorly.
3. Some watermark texts will be recognized, degrade generated srt file.
4. (IMPORTANT) Vertical texts are not supported.
5. Existing srt file will be overwritten, please always make backups.
6. Not much optimization effort was made into this program, it will not utilize all available GPU resources.

Files

mmdocr.exe: Please use this one in actual production environment, this will not show any intermediate processes.
Code is available under src/ folder.