做了什么:
- 包改为分层: tcaptcha/{protocol,telemetry,solvers,ocr,adapters,models,tests}
* protocol/ client.py transport.py proxy.py
* telemetry/ crypto.py values.py key.py opcodes.py
disasm.py program.py layout.py modules.py
* solvers/ text_click.py icon_click.py icon_click_rotate.py slider.py
* ocr/ engines.py slicer.py
* adapters/ one_ms.py
- tools/ 里原本是**包运行时依赖**的 tdc_{disasm,program,layout,modules}.py 并入 telemetry/
(放在 tools/ 属于分层倒置),tools/ 只留 CLI: tdc_key_pure.py / tdc_profile.py / cdp_*.js
- crypto.py 529 -> 340 行: 切除「成功样本模板回填」与「Node 插桩兜底」两条老路;
generate_telemetry 只剩 python_values 一种模式,传别的会直接报错并说明原因
- 求解器改名分清: solver.py -> solvers/icon_click.py(client 实际在用的 YOLO+ResNet18 ONNX),
icon_click_solver.py -> solvers/icon_click_rotate.py(仅测试在用,留作对照)
- pyproject: 去掉 ultralytics/onnx(训练栈已删),补上真正在用的 onnxruntime
删除了什么:
- 雨云整条线 + dist/rainyun_qinglong(104M 旧快照)
- 模板回填: models/success_collect_template.b64、tdc_worker.js、analyze_collect_layout.py
- 「补环境」时代浏览器 JS: chrome_fetch/manual_captcha/probe_captcha/solve_in_browser/tdc_browser
- 旧 VM 探索工具 10 个(tdc_collectgen/tdc_slotmap/tdc_lift/tdc_vm_run/tdc_query/tdc_probe 等)
+ 8 个基于它们的旧 verify 脚本
- 训练/标注/评估: train_classifier/train_detector/prepare_yolo/web_labeler/collector/
siamese/evaluate/benchmark_verify
- tcaptcha/core_js/(60 个分析产物)、TOOLCHAIN_REPORT.md(整篇在描述已删工具)
- 3 个已无任何引用的 PyTorch 权重:icon_classifier.pt 43M / yolo_detector.pt 5.9M /
siamese_matcher.pt 1.6M —— 线上只跑 onnx,tcaptcha/ 117M -> 66M
为什么:
- 项目后面要专门承载「滑块 / 文字点选 / 图标点选」三种求解器;扁平包 + 运行时依赖混在
tools/ + 两条已废弃的 collect 生成路线,会让后续扩展持续踩坑
怎么验证:
- .venv/bin/python -m tcaptcha.selftest 全绿
- scripts/verify/verify_offline.py 逐槽全等 + 分段几何一致
- scripts/verify/verify_values.py 4 份真机明文 152/152 逐槽全等
- scripts/verify/send_test.py --mode python_values 1ms 签到页文字点选 errorCode=0
- 回滚点: tag backup/pre-refactor-20261012
116 lines
4.6 KiB
Python
116 lines
4.6 KiB
Python
# -*- coding: utf-8 -*-
|
|
"""
|
|
滑块拼图求解器 回归测试 (AGENTS.md §5.9)
|
|
|
|
固化的性质 (样本取自 1ms.run aid=195672344 真机页面 2026-10-11 的滑块验证码):
|
|
1. `sprite_region` 能把 CSS 的 background-position/size 正确换算成精灵图像素区域;
|
|
2. ZNCC 峰值落在**真缺口**上, 而不是干扰缺口, 也不是拼图块当前位置;
|
|
3. 通过 y 方向小抖动吸收取整误差, 结果稳定;
|
|
4. `ok` 的判定 (分数 + 与次峰间距) 在该样本上成立。
|
|
|
|
样本: tcaptcha/tests/fixtures/1ms_slider_1_{bg.jpg,sprite.png}
|
|
bg = 页面 `img_index=1` 的背景图 (672x390, 含真缺口 + 干扰缺口)
|
|
sprite = 页面 `img_index=0` 的精灵图 (682x620, 含拖动柄 + 拼图块)
|
|
|
|
DOM 实测口径 (drag_ele.html iframe 内):
|
|
slideBg rect=(10, 68.78) 显示 340 x 242.857 -> scale = 340/672 = 0.50595238
|
|
拼图块元素 rect=(35.2976, 133.53) 显示 60.7143 x 60.7143
|
|
background-position: -70.8333px -247.917px
|
|
background-size: 345.06px 313.69px
|
|
=> 拼图块裁剪图在背景图坐标系里的左上角 = (50.0, 127.97)
|
|
|
|
运行:
|
|
.venv/bin/python -m tcaptcha.tests.test_slider_solver
|
|
"""
|
|
from __future__ import annotations
|
|
|
|
import sys
|
|
from pathlib import Path
|
|
|
|
import cv2
|
|
|
|
ROOT = Path(__file__).resolve().parents[2]
|
|
if str(ROOT) not in sys.path:
|
|
sys.path.insert(0, str(ROOT))
|
|
|
|
from tcaptcha.solvers.slider import SliderSolver, piece_mask, sprite_region # noqa: E402
|
|
|
|
FIX = ROOT / "tcaptcha" / "tests" / "fixtures"
|
|
|
|
# 真机 DOM 口径
|
|
BG_POS = (-70.8333, -247.917)
|
|
BG_SIZE = (345.06, 313.69)
|
|
ELEM_SIZE = (60.7143, 60.7143)
|
|
SLIDE_W = 340.0
|
|
BG_NATURAL_W = 672.0
|
|
SCALE = SLIDE_W / BG_NATURAL_W # 0.50595238
|
|
PIECE_X_IMAGE = 50.0
|
|
PIECE_Y_IMAGE = 128.0
|
|
|
|
EXPECTED_DX_IMAGE = 293.0 # 真缺口 x - 拼图块 x
|
|
EXPECTED_DECOY_X = 489 # 干扰缺口在背景图坐标里的 x
|
|
|
|
_fails: list = []
|
|
|
|
|
|
def check(name: str, cond: bool, detail: str = "") -> None:
|
|
print(f" [{'PASS' if cond else 'FAIL'}] {name}" + (f" {detail}" if detail else ""))
|
|
if not cond:
|
|
_fails.append(name)
|
|
|
|
|
|
def main() -> int:
|
|
bg = cv2.imread(str(FIX / "1ms_slider_1_bg.jpg"))
|
|
sprite = cv2.imread(str(FIX / "1ms_slider_1_sprite.png"), cv2.IMREAD_UNCHANGED)
|
|
if bg is None or sprite is None:
|
|
print("缺少 fixture, 请确认 tcaptcha/tests/fixtures/1ms_slider_1_* 存在")
|
|
return 1
|
|
print(f"bg={bg.shape} sprite={sprite.shape}")
|
|
|
|
print("\n[1] 精灵图区域换算")
|
|
x, y, w, h = sprite_region(BG_POS, BG_SIZE, ELEM_SIZE, sprite.shape[:2])
|
|
check("sprite_region == (140, 490, 120, 120)", (x, y, w, h) == (140, 490, 120, 120), f"got={(x, y, w, h)}")
|
|
|
|
crop = sprite[y:y + h, x:x + w]
|
|
mask = piece_mask(crop)
|
|
check("拼图块掩膜非空且不过分满", 500 < int(mask.sum()) < w * h * 0.9,
|
|
f"mask={int(mask.sum())}/{w * h}")
|
|
|
|
print("\n[2] 缺口定位 (ZNCC)")
|
|
res = SliderSolver().solve(bg, crop, piece_x_image=PIECE_X_IMAGE, scale=SCALE,
|
|
piece_y_image=PIECE_Y_IMAGE)
|
|
print(f" {res.describe()}")
|
|
check("ok", res.ok)
|
|
check("分数 >= 0.65", res.score >= 0.65, f"score={res.score:.3f}")
|
|
check("与次峰间距 >= 0.15", res.margin >= 0.15, f"margin={res.margin:.3f}")
|
|
check("落在真缺口 (误差 <= 4 背景像素)", abs(res.target_x_image - (PIECE_X_IMAGE + EXPECTED_DX_IMAGE)) <= 4,
|
|
f"target_x={res.target_x_image}")
|
|
|
|
print("\n[3] 干扰缺口没有被选中")
|
|
decoy = [x0 for x0, _ in res.candidates if abs(x0 - EXPECTED_DECOY_X) <= 6]
|
|
check("次峰就是干扰缺口", bool(decoy), f"candidates={res.candidates}")
|
|
check("真缺口分数严格高于干扰缺口", res.score > res.runner_up + 0.1,
|
|
f"{res.score:.3f} vs {res.runner_up:.3f}")
|
|
|
|
print("\n[4] 显示坐标换算")
|
|
check("dx_display == dx_image * scale",
|
|
abs(res.dx_display - res.dx_image * SCALE) < 1e-6,
|
|
f"{res.dx_display:.2f} == {res.dx_image:.2f} * {SCALE:.6f}")
|
|
check("dx_display 落在 [140, 160]", 140 <= res.dx_display <= 160, f"{res.dx_display:.2f}")
|
|
|
|
print("\n[5] 拼图块当前位置不是答案 (那里没有洞)")
|
|
near_piece = [s for x0, s in res.candidates if abs(x0 - PIECE_X_IMAGE) <= 8]
|
|
check("拼图块自身位置的分数很低", all(s < 0.3 for s in near_piece),
|
|
f"{near_piece}")
|
|
|
|
print(f"\n{'=' * 62}")
|
|
if _fails:
|
|
print(f"❌ slider_solver 测试失败: {_fails}")
|
|
return 1
|
|
print("✅ slider_solver 测试全部通过")
|
|
return 0
|
|
|
|
|
|
if __name__ == "__main__":
|
|
raise SystemExit(main())
|