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

升级 PaddleOCR 至 3.4.0，移除 PP-OCRv4 模型

Browse Source 
- PaddleOCR 2.10.0 升级至 3.4.0，使用新的 TextDetection API 替代旧版 TextDetector
- V5 模型替换为 PaddleX 格式（inference.pdiparams + inference.yml）
- 移除 PP-OCRv4 模型（backend/models/V4），仅保留 PP-OCRv5
- 字幕检测选项重命名为"快速"和"精准"
- 清理 requirements.txt，移除 11 个未使用的依赖，补充缺失的依赖
- 更新设计素材

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
flavioy
2026-04-07 22:34:36 +08:00
co-authored by Claude Opus 4.6
parent d023ca775c
commit 70e669e9d4
24 changed files with 384 additions and 60 deletions
Download Patch File Download Diff File Expand all files Collapse all files
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.PP_OCRv4_SERVER, OptionsValidator(SubtitleDetectMode), EnumSerializer(SubtitleDetectMode))
subtitleDetectMode = OptionsConfigItem("Main", "SubtitleDetectMode", SubtitleDetectMode.PP_OCRv5_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/ffmpeg/macos/ffmpeg
Regular → Executable
View File
backend/models/V4/ch_det/fs_manifest.csv
-4
View File
@@ -1,4 +0,0 @@
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/models/V5/ch_det/config.json
+111
View File
@@ -0,0 +1,111 @@
{
"Global": {
"model_name": "PP-OCRv5_server_det"
},
"Hpi": {
"backend_configs": {
"paddle_infer": {
"trt_dynamic_shapes": {
"x": [
[
1,
3,
32,
32
],
[
1,
3,
736,
736
],
[
1,
3,
4000,
4000
]
]
}
},
"tensorrt": {
"dynamic_shapes": {
"x": [
[
1,
3,
32,
32
],
[
1,
3,
736,
736
],
[
1,
3,
4000,
4000
]
]
}
}
}
},
"PreProcess": {
"transform_ops": [
{
"DecodeImage": {
"channel_first": false,
"img_mode": "BGR"
}
},
{
"DetLabelEncode": null
},
{
"DetResizeForTest": {
"resize_long": 960
}
},
{
"NormalizeImage": {
"mean": [
0.485,
0.456,
0.406
],
"order": "hwc",
"scale": "1./255.",
"std": [
0.229,
0.224,
0.225
]
}
},
{
"ToCHWImage": null
},
{
"KeepKeys": {
"keep_keys": [
"image",
"shape",
"polys",
"ignore_tags"
]
}
}
]
},
"PostProcess": {
"name": "DBPostProcess",
"thresh": 0.3,
"box_thresh": 0.6,
"max_candidates": 1000,
"unclip_ratio": 1.5
}
}
backend/models/V5/ch_det/fs_manifest.csv
-3
View File
@@ -1,3 +0,0 @@
filename,filesize,encoding,header␍
inference_1.onnx,50000000,,␍
inference_2.onnx,38118768,,␍
1 filename filesize encoding header
2 inference_1.onnx 50000000
3 inference_2.onnx 38118768
backend/models/V5/ch_det/inference.json
+1
View File
File diff suppressed because one or more lines are too long
backend/models/V5/ch_det/inference_1.onnx → backend/models/V5/ch_det/inference.pdiparams
BIN
View File
Binary file not shown.
backend/models/V5/ch_det/inference.yml
+53
View File
@@ -0,0 +1,53 @@
Global:
model_name: PP-OCRv5_server_det
Hpi:
backend_configs:
paddle_infer:
trt_dynamic_shapes: &id001
x:
- - 1
- 3
- 32
- 32
- - 1
- 3
- 736
- 736
- - 1
- 3
- 4000
- 4000
tensorrt:
dynamic_shapes: *id001
PreProcess:
transform_ops:
- DecodeImage:
channel_first: false
img_mode: BGR
- DetLabelEncode: null
- DetResizeForTest:
resize_long: 960
- NormalizeImage:
mean:
- 0.485
- 0.456
- 0.406
order: hwc
scale: 1./255.
std:
- 0.229
- 0.224
- 0.225
- ToCHWImage: null
- KeepKeys:
keep_keys:
- image
- shape
- polys
- ignore_tags
PostProcess:
name: DBPostProcess
thresh: 0.3
box_thresh: 0.6
max_candidates: 1000
unclip_ratio: 1.5
backend/models/V5/ch_det/inference_2.onnx
BIN
View File
Binary file not shown.
backend/models/V5/ch_det_fast/config.json
+111
View File
@@ -0,0 +1,111 @@
{
"Global": {
"model_name": "PP-OCRv5_mobile_det"
},
"Hpi": {
"backend_configs": {
"paddle_infer": {
"trt_dynamic_shapes": {
"x": [
[
1,
3,
32,
32
],
[
1,
3,
736,
736
],
[
1,
3,
4000,
4000
]
]
}
},
"tensorrt": {
"dynamic_shapes": {
"x": [
[
1,
3,
32,
32
],
[
1,
3,
736,
736
],
[
1,
3,
4000,
4000
]
]
}
}
}
},
"PreProcess": {
"transform_ops": [
{
"DecodeImage": {
"channel_first": false,
"img_mode": "BGR"
}
},
{
"DetLabelEncode": null
},
{
"DetResizeForTest": {
"resize_long": 960
}
},
{
"NormalizeImage": {
"mean": [
0.485,
0.456,
0.406
],
"order": "hwc",
"scale": "1./255.",
"std": [
0.229,
0.224,
0.225
]
}
},
{
"ToCHWImage": null
},
{
"KeepKeys": {
"keep_keys": [
"image",
"shape",
"polys",
"ignore_tags"
]
}
}
]
},
"PostProcess": {
"name": "DBPostProcess",
"thresh": 0.3,
"box_thresh": 0.6,
"max_candidates": 1000,
"unclip_ratio": 1.5
}
}
backend/models/V5/ch_det_fast/inference.json
+1
View File
File diff suppressed because one or more lines are too long
backend/models/V5/ch_det_fast/inference.onnx → backend/models/V5/ch_det_fast/inference.pdiparams
BIN
View File
Binary file not shown.
backend/models/V5/ch_det_fast/inference.yml
+53
View File
@@ -0,0 +1,53 @@
Global:
model_name: PP-OCRv5_mobile_det
Hpi:
backend_configs:
paddle_infer:
trt_dynamic_shapes: &id001
x:
- - 1
- 3
- 32
- 32
- - 1
- 3
- 736
- 736
- - 1
- 3
- 4000
- 4000
tensorrt:
dynamic_shapes: *id001
PreProcess:
transform_ops:
- DecodeImage:
channel_first: false
img_mode: BGR
- DetLabelEncode: null
- DetResizeForTest:
resize_long: 960
- NormalizeImage:
mean:
- 0.485
- 0.456
- 0.406
order: hwc
scale: 1./255.
std:
- 0.229
- 0.224
- 0.225
- ToCHWImage: null
- KeepKeys:
keep_keys:
- image
- shape
- polys
- ignore_tags
PostProcess:
name: DBPostProcess
thresh: 0.3
box_thresh: 0.6
max_candidates: 1000
unclip_ratio: 1.5
backend/tools/constant.py
+2 -4
View File
@@ -16,7 +16,5 @@ class SubtitleDetectMode(Enum):
"""
字幕检测算法枚举
"""
PP_OCRv4_SERVER = "PP-OCRv4-Server"
PP_OCRv4_MOBILE = "PP-OCRv4-Mobile"
PP_OCRv5_SERVER = "PP-OCRv5-Server"
PP_OCRv5_MOBILE = "PP-OCRv5-Mobile"
PP_OCRv5_MOBILE = "快速"
PP_OCRv5_SERVER = "精准"
backend/tools/model_config.py
+6 -6
View File
@@ -3,6 +3,11 @@ from backend.config import config, BASE_DIR
from backend.tools.common_tools import merge_big_file_if_not_exists
from backend.tools.constant import SubtitleDetectMode
_MODEL_NAME_MAP = {
SubtitleDetectMode.PP_OCRv5_MOBILE: "PP-OCRv5_mobile_det",
SubtitleDetectMode.PP_OCRv5_SERVER: "PP-OCRv5_server_det",
}
class ModelConfig:
def __init__(self):
self.LAMA_MODEL_DIR = os.path.join(BASE_DIR, 'models', 'big-lama')
@@ -13,14 +18,9 @@ class ModelConfig:
self.DET_MODEL_DIR = os.path.join(BASE_DIR,'models', 'V5', 'ch_det_fast')
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}")
self.DET_MODEL_NAME = _MODEL_NAME_MAP[config.subtitleDetectMode.value]
merge_big_file_if_not_exists(self.LAMA_MODEL_DIR, 'bit-lama.pt')
merge_big_file_if_not_exists(self.PROPAINTER_MODEL_DIR, 'ProPainter.pth')
merge_big_file_if_not_exists(self.DET_MODEL_DIR, 'inference.onnx')
backend/tools/subtitle_detect.py
+25 -25
View File
@@ -1,4 +1,3 @@
import os
import sys
from functools import cached_property
@@ -27,36 +26,37 @@ class SubtitleDetect:
def text_detector(self):
import paddle
paddle.disable_signal_handler()
from paddleocr.tools.infer import utility
from paddleocr.tools.infer.predict_det import TextDetector
from paddleocr import TextDetection
hardware_accelerator = HardwareAccelerator.instance()
onnx_providers = hardware_accelerator.onnx_providers
model_config = ModelConfig()
parser = utility.init_args()
args = parser.parse_args([])
args.det_algorithm = 'DB'
args.det_model_dir = os.path.join(model_config.DET_MODEL_DIR, 'inference.onnx') if len(onnx_providers) > 0 else model_config.DET_MODEL_DIR
args.use_gpu=False
args.use_onnx=len(onnx_providers) > 0
args.onnx_providers=onnx_providers
return TextDetector(args)
return TextDetection(
model_name=model_config.DET_MODEL_NAME,
model_dir=model_config.DET_MODEL_DIR,
device="cpu",
enable_hpi=len(onnx_providers) > 0,
)
def detect_subtitle(self, img):
temp_list = []
dt_boxes, elapse = self.text_detector(img)
coordinate_list = get_coordinates(dt_boxes.tolist())
if coordinate_list:
for coordinate in coordinate_list:
xmin, xmax, ymin, ymax = coordinate
if self.sub_areas is not None and len(self.sub_areas) > 0:
for sub_area in self.sub_areas:
s_ymin, s_ymax, s_xmin, s_xmax = sub_area
if (s_xmin <= xmin and xmax <= s_xmax
and s_ymin <= ymin
and ymax <= s_ymax):
temp_list.append((xmin, xmax, ymin, ymax))
else:
temp_list.append((xmin, xmax, ymin, ymax))
results = self.text_detector.predict(img)
for res in results:
dt_polys = res['dt_polys']
if dt_polys is None or len(dt_polys) == 0:
continue
coordinate_list = get_coordinates(dt_polys.tolist())
if coordinate_list:
for coordinate in coordinate_list:
xmin, xmax, ymin, ymax = coordinate
if self.sub_areas is not None and len(self.sub_areas) > 0:
for sub_area in self.sub_areas:
s_ymin, s_ymax, s_xmin, s_xmax = sub_area
if (s_xmin <= xmin and xmax <= s_xmax
and s_ymin <= ymin
and ymax <= s_ymax):
temp_list.append((xmin, xmax, ymin, ymax))
else:
temp_list.append((xmin, xmax, ymin, ymax))
return temp_list
def find_subtitle_frame_no(self, sub_remover=None):
design/bg.png
BIN
View File
Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 405 KiB

design/icon_1024.PNG
BIN
View File
Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1.2 MiB

design/vsr.ico
BIN
View File
Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 66 KiB

After

Width:  |  Height:  |  Size: 26 KiB

requirements.txt
+20 -17
View File
@@ -1,25 +1,28 @@
albumentations~=1.4.10
filesplit==3.0.2
opencv-python==4.11.0.86
scikit-image==0.25.2
imgaug==0.4.0
pyclipper==1.3.0.post6
lmdb==1.6.2
PyYAML==6.0.2
omegaconf==2.3.0
tqdm==4.67.1
pyside6-fluent-widgets==1.7.7
easydict==1.13
scikit-learn==1.6.1
pandas==2.2.3
webdataset==0.2.111
numpy==2.2.5
protobuf==6.30.2
av==14.3.0
av==17.0.0
einops==0.8.1
paddleocr==2.10.0
onnxruntime-gpu==1.20.1; sys_platform == 'linux'
paddleocr==3.4.0
matplotlib
Pillow
scipy
requests
darkdetect
je-showinfilemanager==1.1.6a4
# torch and torchvision are installed separately (see Dockerfile for platform-specific versions)
# Training only
#matplotlib
#lpips
#tensorboardX
# Hardware acceleration
#onnxruntime-gpu==1.20.1; sys_platform == 'linux'
# do not upgrade, windows 10/11 compatable issues, check hardware_acclerator.py:100
onnxruntime-directml==1.20.1; sys_platform == 'win32'
onnxruntime==1.22.0; sys_platform == 'darwin'
je-showinfilemanager==1.1.6a4
#onnxruntime-directml==1.20.1; sys_platform == 'win32'
#onnxruntime==1.22.0; sys_platform == 'darwin'