Expand 3D captcha into three subtypes: 3d_text, 3d_rotate, 3d_slider
Split the single "3d" captcha type into three independent expert models: - 3d_text: 3D perspective text OCR (renamed from old "3d", CTC-based ThreeDCNN) - 3d_rotate: rotation angle regression (new RegressionCNN, circular loss) - 3d_slider: slider offset regression (new RegressionCNN, SmoothL1 loss) CAPTCHA_TYPES expanded from 3 to 5 classes. Classifier samples updated to 50000 (10000 per class). New generators, model, dataset, training utilities, and full pipeline/export/CLI support for all subtypes. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
83
cli.py
83
cli.py
@@ -3,6 +3,9 @@ CaptchaBreaker 命令行入口
|
||||
|
||||
用法:
|
||||
python cli.py generate --type normal --num 60000
|
||||
python cli.py generate --type 3d_text --num 80000
|
||||
python cli.py generate --type 3d_rotate --num 60000
|
||||
python cli.py generate --type 3d_slider --num 60000
|
||||
python cli.py train --model normal
|
||||
python cli.py train --all
|
||||
python cli.py export --all
|
||||
@@ -20,15 +23,21 @@ from pathlib import Path
|
||||
def cmd_generate(args):
|
||||
"""生成训练数据。"""
|
||||
from config import (
|
||||
SYNTHETIC_NORMAL_DIR, SYNTHETIC_MATH_DIR, SYNTHETIC_3D_DIR,
|
||||
SYNTHETIC_NORMAL_DIR, SYNTHETIC_MATH_DIR,
|
||||
SYNTHETIC_3D_TEXT_DIR, SYNTHETIC_3D_ROTATE_DIR, SYNTHETIC_3D_SLIDER_DIR,
|
||||
CLASSIFIER_DIR, TRAIN_CONFIG, CAPTCHA_TYPES, NUM_CAPTCHA_TYPES,
|
||||
)
|
||||
from generators import NormalCaptchaGenerator, MathCaptchaGenerator, ThreeDCaptchaGenerator
|
||||
from generators import (
|
||||
NormalCaptchaGenerator, MathCaptchaGenerator, ThreeDCaptchaGenerator,
|
||||
ThreeDRotateGenerator, ThreeDSliderGenerator,
|
||||
)
|
||||
|
||||
gen_map = {
|
||||
"normal": (NormalCaptchaGenerator, SYNTHETIC_NORMAL_DIR),
|
||||
"math": (MathCaptchaGenerator, SYNTHETIC_MATH_DIR),
|
||||
"3d": (ThreeDCaptchaGenerator, SYNTHETIC_3D_DIR),
|
||||
"3d_text": (ThreeDCaptchaGenerator, SYNTHETIC_3D_TEXT_DIR),
|
||||
"3d_rotate": (ThreeDRotateGenerator, SYNTHETIC_3D_ROTATE_DIR),
|
||||
"3d_slider": (ThreeDSliderGenerator, SYNTHETIC_3D_SLIDER_DIR),
|
||||
}
|
||||
|
||||
captcha_type = args.type
|
||||
@@ -50,25 +59,31 @@ def cmd_generate(args):
|
||||
gen = gen_cls()
|
||||
gen.generate_dataset(num, str(out_dir))
|
||||
else:
|
||||
print(f"未知类型: {captcha_type} 可选: normal, math, 3d, classifier")
|
||||
valid = ", ".join(list(gen_map.keys()) + ["classifier"])
|
||||
print(f"未知类型: {captcha_type} 可选: {valid}")
|
||||
sys.exit(1)
|
||||
|
||||
|
||||
def cmd_train(args):
|
||||
"""训练模型。"""
|
||||
if args.all:
|
||||
# 按依赖顺序: normal → math → 3d → classifier
|
||||
print("按顺序训练全部模型: normal → math → 3d → classifier\n")
|
||||
print("按顺序训练全部模型: normal → math → 3d_text → 3d_rotate → 3d_slider → classifier\n")
|
||||
from training.train_normal import main as train_normal
|
||||
from training.train_math import main as train_math
|
||||
from training.train_3d import main as train_3d
|
||||
from training.train_3d_text import main as train_3d_text
|
||||
from training.train_3d_rotate import main as train_3d_rotate
|
||||
from training.train_3d_slider import main as train_3d_slider
|
||||
from training.train_classifier import main as train_classifier
|
||||
|
||||
train_normal()
|
||||
print("\n")
|
||||
train_math()
|
||||
print("\n")
|
||||
train_3d()
|
||||
train_3d_text()
|
||||
print("\n")
|
||||
train_3d_rotate()
|
||||
print("\n")
|
||||
train_3d_slider()
|
||||
print("\n")
|
||||
train_classifier()
|
||||
return
|
||||
@@ -78,12 +93,16 @@ def cmd_train(args):
|
||||
from training.train_normal import main as train_fn
|
||||
elif model == "math":
|
||||
from training.train_math import main as train_fn
|
||||
elif model == "3d":
|
||||
from training.train_3d import main as train_fn
|
||||
elif model == "3d_text":
|
||||
from training.train_3d_text import main as train_fn
|
||||
elif model == "3d_rotate":
|
||||
from training.train_3d_rotate import main as train_fn
|
||||
elif model == "3d_slider":
|
||||
from training.train_3d_slider import main as train_fn
|
||||
elif model == "classifier":
|
||||
from training.train_classifier import main as train_fn
|
||||
else:
|
||||
print(f"未知模型: {model} 可选: normal, math, 3d, classifier")
|
||||
print(f"未知模型: {model} 可选: normal, math, 3d_text, 3d_rotate, 3d_slider, classifier")
|
||||
sys.exit(1)
|
||||
|
||||
train_fn()
|
||||
@@ -96,7 +115,14 @@ def cmd_export(args):
|
||||
if args.all:
|
||||
export_all()
|
||||
elif args.model:
|
||||
_load_and_export(args.model)
|
||||
# 别名映射
|
||||
alias = {
|
||||
"3d_text": "threed_text",
|
||||
"3d_rotate": "threed_rotate",
|
||||
"3d_slider": "threed_slider",
|
||||
}
|
||||
name = alias.get(args.model, args.model)
|
||||
_load_and_export(name)
|
||||
else:
|
||||
print("请指定 --all 或 --model <name>")
|
||||
sys.exit(1)
|
||||
@@ -137,19 +163,19 @@ def cmd_predict_dir(args):
|
||||
sys.exit(1)
|
||||
|
||||
print(f"批量识别: {len(images)} 张图片\n")
|
||||
print(f"{'文件名':<30} {'类型':<8} {'结果':<15} {'耗时(ms)':>8}")
|
||||
print("-" * 65)
|
||||
print(f"{'文件名':<30} {'类型':<10} {'结果':<15} {'耗时(ms)':>8}")
|
||||
print("-" * 67)
|
||||
|
||||
total_ms = 0.0
|
||||
for img_path in images:
|
||||
result = pipeline.solve(str(img_path), captcha_type=args.type)
|
||||
total_ms += result["time_ms"]
|
||||
print(
|
||||
f"{img_path.name:<30} {result['type']:<8} "
|
||||
f"{img_path.name:<30} {result['type']:<10} "
|
||||
f"{result['result']:<15} {result['time_ms']:>8.1f}"
|
||||
)
|
||||
|
||||
print("-" * 65)
|
||||
print("-" * 67)
|
||||
print(f"总计: {len(images)} 张 平均: {total_ms / len(images):.1f} ms 总耗时: {total_ms:.1f} ms")
|
||||
|
||||
|
||||
@@ -178,28 +204,43 @@ def main():
|
||||
|
||||
# ---- generate ----
|
||||
p_gen = subparsers.add_parser("generate", help="生成训练数据")
|
||||
p_gen.add_argument("--type", required=True, help="验证码类型: normal, math, 3d, classifier")
|
||||
p_gen.add_argument(
|
||||
"--type", required=True,
|
||||
help="验证码类型: normal, math, 3d_text, 3d_rotate, 3d_slider, classifier",
|
||||
)
|
||||
p_gen.add_argument("--num", type=int, required=True, help="生成数量")
|
||||
|
||||
# ---- train ----
|
||||
p_train = subparsers.add_parser("train", help="训练模型")
|
||||
p_train.add_argument("--model", help="模型名: normal, math, 3d, classifier")
|
||||
p_train.add_argument(
|
||||
"--model",
|
||||
help="模型名: normal, math, 3d_text, 3d_rotate, 3d_slider, classifier",
|
||||
)
|
||||
p_train.add_argument("--all", action="store_true", help="按依赖顺序训练全部模型")
|
||||
|
||||
# ---- export ----
|
||||
p_export = subparsers.add_parser("export", help="导出 ONNX 模型")
|
||||
p_export.add_argument("--model", help="模型名: normal, math, 3d, classifier, threed")
|
||||
p_export.add_argument(
|
||||
"--model",
|
||||
help="模型名: normal, math, 3d_text, 3d_rotate, 3d_slider, classifier",
|
||||
)
|
||||
p_export.add_argument("--all", action="store_true", help="导出全部模型")
|
||||
|
||||
# ---- predict ----
|
||||
p_pred = subparsers.add_parser("predict", help="识别单张验证码")
|
||||
p_pred.add_argument("image", help="图片路径")
|
||||
p_pred.add_argument("--type", default=None, help="指定类型跳过分类: normal, math, 3d")
|
||||
p_pred.add_argument(
|
||||
"--type", default=None,
|
||||
help="指定类型跳过分类: normal, math, 3d_text, 3d_rotate, 3d_slider",
|
||||
)
|
||||
|
||||
# ---- predict-dir ----
|
||||
p_pdir = subparsers.add_parser("predict-dir", help="批量识别目录中的验证码")
|
||||
p_pdir.add_argument("directory", help="图片目录路径")
|
||||
p_pdir.add_argument("--type", default=None, help="指定类型跳过分类: normal, math, 3d")
|
||||
p_pdir.add_argument(
|
||||
"--type", default=None,
|
||||
help="指定类型跳过分类: normal, math, 3d_text, 3d_rotate, 3d_slider",
|
||||
)
|
||||
|
||||
# ---- serve ----
|
||||
p_serve = subparsers.add_parser("serve", help="启动 HTTP 识别服务")
|
||||
|
||||
Reference in New Issue
Block a user