admin/video-subtitle-remover
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

同时支持PP-OCRv4 PP-OCRv5模型

Browse Source
This commit is contained in:
Jason
2025-05-22 19:52:36 +08:00
parent afcc93b9cc
commit 50c89c1217
19 changed files with 34 additions and 52 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+1 -1
View File
@@ -201,7 +201,7 @@ cd <源码所在目录>
##### (4) macOS 运行 (Apple Silicon)
- 适用于 macOS (Apple Silicon) 设备
- macOS (Intel) 请使用CPU, 强行使用GPU只会更慢
- macOS (Apple Silicon)上字幕检测精准模式的模型(ch_det_fast/model.onnx)精度似乎不太理想, 推荐使用快速模式
- macOS (Apple Silicon)上字幕检测PP-OCRv4-Server模型精度似乎不太理想, 推荐使用其他模型
```shell
pip install paddlepaddle==3.0.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
pip install torch==2.7.0 torchvision==0.22.0
README_en.md
+1 -1
View File
@@ -202,7 +202,7 @@ This project supports four running modes: CUDA (NVIDIA GPU acceleration), CPU (n
##### (4) Running on macOS (Apple Silicon)
- Suitable for macOS (Apple Silicon) devices
- For macOS (Intel), please use the CPU mode. Forcing GPU usage will only be slower.
- On macOS (Apple Silicon), the accuracy of the subtitle detection precision mode model (ch_det_fast/model.onnx) does not seem to be ideal. It is recommended to use the fast mode instead.
- On macOS (Apple Silicon), the accuracy of the PP-OCRv4-Server model for subtitle detection seems suboptimal. We recommend using an alternative model.
```shell
pip install paddlepaddle==3.0.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
pip install torch==2.7.0 torchvision==0.22.0
backend/config.py
+1 -1
View File
@@ -52,7 +52,7 @@ class Config(QConfig):
# 【设置inpaint算法】
inpaintMode = OptionsConfigItem("Main", "InpaintMode", InpaintMode.STTN_AUTO, OptionsValidator(InpaintMode), EnumSerializer(InpaintMode))
subtitleDetectMode = OptionsConfigItem("Main", "SubtitleDetectMode", SubtitleDetectMode.Accurate, OptionsValidator(SubtitleDetectMode), EnumSerializer(SubtitleDetectMode))
subtitleDetectMode = OptionsConfigItem("Main", "SubtitleDetectMode", SubtitleDetectMode.PP_OCRv4_SERVER, OptionsValidator(SubtitleDetectMode), EnumSerializer(SubtitleDetectMode))
# 【设置像素点偏差】
# 用于判断是不是非字幕区域(一般认为字幕文本框的长度是要大于宽度的，如果字幕框的高大于宽，且大于的幅度超过指定像素点大小，则认为是错误检测)
backend/interface/ch.ini
+2 -6
View File
@@ -76,7 +76,6 @@ DeleteABSection = 删除当前处理区块
DeleteSelection = 删除当前激活选区
[Main]
SubtitleDetectionAcceleratorON = 字幕检测使用{}进行加速
OnnxExectionProviderNotSupportedSkipped = ONNX 执行提供程序: {} 不支持，已跳过。
OnnxExecutionProviderDetected=检测到 ONNX 执行提供程序: {}
OnnxRuntimeNotInstall = ONNX 运行环境未安装，已跳过。
@@ -85,7 +84,8 @@ DirectMLWarning = 警告: DirectML 加速仅适用于 STTN 模型，其他模型
ProcessingStartFindingSubtitles = [处理中] 开始查找字幕...
FinishedFindingSubtitles = [结束] 查找字幕完成...
ProcessingStartRemovingSubtitles = [处理中] 开始移除字幕...
UseModel = 去除字幕使用模型: {}
SubtitleRemoverModel = 去除字幕模型: {}
SubtitleDetectionModel = 字幕检测模型: {}
FullScreenProcessingNote = 未设置字幕区域，将对全屏进行处理，最终效果可能不理想
ReadFileFailed = 读取文件 {} 失败
FinishedProcessing = [完成] 字幕移除成功, 文件已保存到 {}
@@ -124,10 +124,6 @@ LAMA = LAMA
ProPainter = ProPainter
OpenCV = OpenCV
[SubtitleDetectMode]
Fast = 快速
Accurate = 精准
[InterfaceLanguage]
ChineseSimplified = 简体中文
ChineseTraditional = 繁體中文
backend/interface/chinese_cht.ini
+2 -6
View File
@@ -76,7 +76,6 @@ DeleteABSection = 刪除當前處理區塊
DeleteSelection = 刪除當前激活選區
[Main]
SubtitleDetectionAcceleratorON = 字幕檢測使用{}進行加速
OnnxExectionProviderNotSupportedSkipped = ONNX 執行提供者: {} 不支援，已略過。
OnnxExecutionProviderDetected = 偵測到 ONNX 執行提供者: {}
OnnxRuntimeNotInstall = ONNX 執行環境未安裝，已略過。
@@ -85,7 +84,8 @@ DirectMLWarning = 警告：DirectML 加速僅適用於 STTN 模型，其他模
ProcessingStartFindingSubtitles = [處理中] 開始搜尋字幕...
FinishedFindingSubtitles = [結束] 搜尋字幕完成...
ProcessingStartRemovingSubtitles = [處理中] 開始移除字幕...
UseModel = 去除字幕使用模型: {}
SubtitleRemoverModel = 去除字幕模型: {}
SubtitleDetectionModel = 字幕檢測模型: {}
FullScreenProcessingNote = 未設定字幕區域，將對全螢幕進行處理，最終效果可能不理想
ReadFileFailed = 讀取檔案 {} 失敗
FinishedProcessing = [完成] 字幕移除成功，檔案已儲存至 {}
@@ -124,10 +124,6 @@ LAMA = LAMA
ProPainter = ProPainter
OpenCV = OpenCV
[SubtitleDetectMode]
Fast = 快速
Accurate = 精準
[InterfaceLanguage]
ChineseSimplified = 简体中文
ChineseTraditional = 繁體中文
backend/interface/en.ini
+2 -6
View File
@@ -76,7 +76,6 @@ DeleteABSection = Delete Section
DeleteSelection = Delete Selection
[Main]
SubtitleDetectionAcceleratorON = Subtitle detection accelerated with {}
OnnxExectionProviderNotSupportedSkipped = ONNX provider: {} not supported, skipped.
OnnxExecutionProviderDetected = Detected ONNX provider: {}
OnnxRuntimeNotInstall = ONNX runtime not installed, skipped.
@@ -85,7 +84,8 @@ DirectMLWarning = Warning: DirectML acceleration only works with STTN model.
ProcessingStartFindingSubtitles = [Processing] Detecting subtitles...
FinishedFindingSubtitles = [Complete] Subtitle detection finished.
ProcessingStartRemovingSubtitles = [Processing] Removing subtitles...
UseModel = Use model for subtitle removal: {}
SubtitleRemoverModel = Subtitle removal model: {}
SubtitleDetectionModel = Subtitle detection model: {}
FullScreenProcessingNote = Processing full screen (no area selected). Quality may vary.
ReadFileFailed = Failed to read file: {}
FinishedProcessing = [Complete] Subtitles removed. Output saved to: {}
@@ -124,10 +124,6 @@ LAMA = LAMA
ProPainter = ProPainter
OpenCV = OpenCV
[SubtitleDetectMode]
Fast = Fast
Accurate = Accurate
[InterfaceLanguage]
ChineseSimplified = 简体中文
ChineseTraditional = 繁體中文
backend/interface/es.ini
+2 -6
View File
@@ -76,7 +76,6 @@ DeleteABSection = Eliminar sección
DeleteSelection = Eliminar selección
[Main]
SubtitleDetectionAcceleratorON = Detección de subtítulos acelerada con {}
OnnxExectionProviderNotSupportedSkipped = Proveedor ONNX: {} no soportado, omitido.
OnnxExecutionProviderDetected = Proveedor ONNX detectado: {}
OnnxRuntimeNotInstall = Entorno ONNX no instalado, omitido.
@@ -85,7 +84,8 @@ DirectMLWarning = Advertencia: Aceleración DirectML solo funciona con modelo ST
ProcessingStartFindingSubtitles = [Procesando] Detectando subtítulos...
FinishedFindingSubtitles = [Completo] Detección finalizada.
ProcessingStartRemovingSubtitles = [Procesando] Eliminando subtítulos...
UseModel = Usar modelo para eliminar subtítulos: {}
SubtitleRemoverModel = Modelo de eliminación de subtítulos: {}
SubtitleDetectionModel = Modelo de detección de subtítulos: {}
FullScreenProcessingNote = Procesando pantalla completa (área no seleccionada).
ReadFileFailed = Error al leer archivo: {}
FinishedProcessing = [Completo] Subtítulos eliminados. Guardado en: {}
@@ -124,10 +124,6 @@ LAMA = LAMA
ProPainter = ProPainter
OpenCV = OpenCV
[SubtitleDetectMode]
Fast = Rápido
Accurate = Preciso
[InterfaceLanguage]
ChineseSimplified = 简体中文
ChineseTraditional = 繁體中文
backend/interface/japan.ini
+2 -6
View File
@@ -76,7 +76,6 @@ DeleteABSection = 現在の処理区間を削除
DeleteSelection = 現在のアクティブ選択範囲を削除
[Main]
SubtitleDetectionAcceleratorON = 字幕検出を{}で加速
OnnxExectionProviderNotSupportedSkipped = ONNXプロバイダ: {} 非対応
OnnxExecutionProviderDetected=ONNXプロバイダ検出: {}
OnnxRuntimeNotInstall = ONNXランタイム未インストール
@@ -85,7 +84,8 @@ DirectMLWarning = 警告: DirectML加速はSTTNモデルのみ
ProcessingStartFindingSubtitles = [処理中] 字幕検索開始...
FinishedFindingSubtitles = [完了] 字幕検索終了
ProcessingStartRemovingSubtitles = [処理中] 字幕削除開始...
UseModel = 字幕除去用モデル: {}
SubtitleRemoverModel = 字幕除去モデル: {}
SubtitleDetectionModel = 字幕検出モデル: {}
FullScreenProcessingNote = 全画面処理（領域未選択）
ReadFileFailed = ファイル読み込み失敗: {}
FinishedProcessing = [完了] 字幕削除成功。保存先: {}
@@ -124,10 +124,6 @@ LAMA = LAMA
ProPainter = ProPainter
OpenCV = OpenCV
[SubtitleDetectMode]
Fast = 高速
Accurate = 高精度
[InterfaceLanguage]
ChineseSimplified = 简体中文
ChineseTraditional = 繁體中文
backend/interface/ko.ini
+2 -6
View File
@@ -76,7 +76,6 @@ DeleteABSection = 현재 처리 구간 삭제
DeleteSelection = 현재 활성 선택 영역 삭제
[Main]
SubtitleDetectionAcceleratorON = 자막 감지 {} 가속 사용
OnnxExectionProviderNotSupportedSkipped = ONNX 공급자: {} 지원 안됨
OnnxExecutionProviderDetected=ONNX 공급자 감지: {}
OnnxRuntimeNotInstall = ONNX 런타임 미설치
@@ -85,7 +84,8 @@ DirectMLWarning = 경고: DirectML 가속은 STTN 모델 전용
ProcessingStartFindingSubtitles = [진행중] 자막 검색 시작...
FinishedFindingSubtitles = [완료] 자막 검색 완료
ProcessingStartRemovingSubtitles = [진행중] 자막 제거 시작...
UseModel = 자막 제거 모델 사용: {}
SubtitleRemoverModel = 자막 제거 모델: {}
SubtitleDetectionModel = 자막 감지 모델: {}
FullScreenProcessingNote = 전체 화면 처리 (영역 미선택)
ReadFileFailed = 파일 읽기 실패: {}
FinishedProcessing = [완료] 자막 제거 완료. 저장 위치: {}
@@ -124,10 +124,6 @@ LAMA = LAMA
ProPainter = ProPainter
OpenCV = OpenCV
[SubtitleDetectMode]
Fast = 빠름
Accurate = 정확
[InterfaceLanguage]
ChineseSimplified = 简体中文
ChineseTraditional = 繁體中文
backend/interface/vi.ini
+2 -6
View File
@@ -76,7 +76,6 @@ DeleteABSection = Xóa vùng xử lý hiện tại
DeleteSelection = Xóa vùng chọn hiện tại
[Main]
SubtitleDetectionAcceleratorON = Phát hiện phụ đề được tăng tốc bằng {}
OnnxExectionProviderNotSupportedSkipped = ONNX provider: {} không hỗ trợ
OnnxExecutionProviderDetected= Phát hiện ONNX provider: {}
OnnxRuntimeNotInstall = Chưa cài đặt ONNX Runtime
@@ -85,7 +84,8 @@ DirectMLWarning = Cảnh báo: DirectML chỉ hỗ trợ STTN
ProcessingStartFindingSubtitles = [Đang xử lý] Bắt đầu tìm phụ đề...
FinishedFindingSubtitles = [Hoàn thành] Tìm phụ đề xong
ProcessingStartRemovingSubtitles = [Đang xử lý] Bắt đầu xóa phụ đề...
UseModel = Sử dụng mô hình xóa phụ đề: {}
SubtitleRemoverModel = Mô hình xóa phụ đề: {}
SubtitleDetectionModel = Mô hình phát hiện phụ đề: {}
FullScreenProcessingNote = Xử lý toàn màn hình (không chọn vùng)
ReadFileFailed = Lỗi đọc file: {}
FinishedProcessing = [Hoàn thành] Xóa phụ đề thành công, lưu tại: {}
@@ -124,10 +124,6 @@ LAMA = LAMA
ProPainter = ProPainter
OpenCV = OpenCV
[SubtitleDetectMode]
Fast = Nhanh
Accurate = Chính xác
[InterfaceLanguage]
ChineseSimplified = 简体中文
ChineseTraditional = 繁體中文
backend/main.py
+2 -2
View File
@@ -334,7 +334,6 @@ class SubtitleRemover:
# 如果使用GPU加速，则打印GPU加速提示
if self.hardware_accelerator.has_accelerator():
accelerator_name = self.hardware_accelerator.accelerator_name
self.append_output(tr['Main']['SubtitleDetectionAcceleratorON'].format(accelerator_name))
if accelerator_name == 'DirectML' and config.inpaintMode.value not in [InpaintMode.STTN_AUTO, InpaintMode.STTN_DET]:
self.append_output(tr['Main']['DirectMLWarning'])
os.makedirs(os.path.dirname(self.video_out_path), exist_ok=True)
@@ -401,7 +400,8 @@ class SubtitleRemover:
model_device = 'DirectML'
if self.hardware_accelerator.has_cuda() or self.hardware_accelerator.has_mps():
model_device = accelerator_name
self.append_output(tr['Main']['UseModel'].format(f"{model_friendly_name} ({model_device})"))
self.append_output(tr['Main']['SubtitleRemoverModel'].format(f"{model_friendly_name} ({model_device})"))
self.append_output(tr['Main']['SubtitleDetectionModel'].format(f"{config.subtitleDetectMode.value.value} ({", ".join(self.hardware_accelerator.onnx_providers)})"))
def merge_audio_to_video(self):
# 创建音频临时对象，windows下delete=True会有permission denied的报错
backend/models/V4/ch_det/fs_manifest.csv
+4
View File
@@ -0,0 +1,4 @@
filename,filesize,encoding,header␍
inference_1.onnx,50000000,,␍
inference_2.onnx,50000000,,␍
inference_3.onnx,13443410,,␍
1 filename filesize encoding header
2 inference_1.onnx 50000000
3 inference_2.onnx 50000000
4 inference_3.onnx 13443410
backend/models/V4/ch_det/inference_1.onnx
BIN
View File
Binary file not shown.
backend/models/V4/ch_det/inference_2.onnx
BIN
View File
Binary file not shown.
backend/models/V4/ch_det/inference_3.onnx
BIN
View File
Binary file not shown.
backend/models/V4/ch_det_fast/inference.onnx
BIN
View File
Binary file not shown.
backend/tools/constant.py
+4 -2
View File
@@ -16,5 +16,7 @@ class SubtitleDetectMode(Enum):
"""
字幕检测算法枚举
"""
Fast = 0
Accurate = 1
PP_OCRv4_SERVER = "PP-OCRv4-Server"
PP_OCRv4_MOBILE = "PP-OCRv4-Mobile"
PP_OCRv5_SERVER = "PP-OCRv5-Server"
PP_OCRv5_MOBILE = "PP-OCRv5-Mobile"
backend/tools/model_config.py
+6 -2
View File
@@ -9,10 +9,14 @@ class ModelConfig:
self.STTN_AUTO_MODEL_PATH = os.path.join(BASE_DIR, 'models', 'sttn-auto', 'infer_model.pth')
self.STTN_DET_MODEL_PATH = os.path.join(BASE_DIR, 'models', 'sttn-det', 'sttn.pth')
self.PROPAINTER_MODEL_DIR = os.path.join(BASE_DIR,'models', 'propainter')
if config.subtitleDetectMode.value == SubtitleDetectMode.Fast:
if config.subtitleDetectMode.value == SubtitleDetectMode.PP_OCRv5_MOBILE:
self.DET_MODEL_DIR = os.path.join(BASE_DIR,'models', 'V5', 'ch_det_fast')
elif config.subtitleDetectMode.value == SubtitleDetectMode.Accurate:
elif config.subtitleDetectMode.value == SubtitleDetectMode.PP_OCRv5_SERVER:
self.DET_MODEL_DIR = os.path.join(BASE_DIR, 'models', 'V5', 'ch_det')
elif config.subtitleDetectMode.value == SubtitleDetectMode.PP_OCRv4_MOBILE:
self.DET_MODEL_DIR = os.path.join(BASE_DIR,'models', 'V4', 'ch_det_fast')
elif config.subtitleDetectMode.value == SubtitleDetectMode.PP_OCRv4_SERVER:
self.DET_MODEL_DIR = os.path.join(BASE_DIR, 'models', 'V4', 'ch_det')
else:
raise ValueError(f"Invalid subtitle detect mode: {config.subtitleDetectMode.value}")
ui/setting_interface.py
+1 -1
View File
@@ -43,7 +43,7 @@ class SettingInterface(QtWidgets.QVBoxLayout):
title=tr["SubtitleExtractorGUI"]["SubtitleDetectMode"],
content="",
parent=parent,
texts=[list(tr['SubtitleDetectMode'].values())[i] for i,_ in enumerate(config.subtitleDetectMode.validator.options)],
texts=[mode.value for mode in config.subtitleDetectMode.validator.options],
)
self.addWidget(self.subtitle_detect_model_combo)