Spaces:
Sleeping
Sleeping
File size: 21,294 Bytes
e40db0e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 | # -*- coding: utf-8 -*-
"""
数据验证器
提供数据质量检查、格式验证和完整性检测功能
"""
import json
import logging
import re
from collections import Counter
from pathlib import Path
from typing import Any, Dict, List, Union
from PIL import Image
from tqdm.auto import tqdm
logger = logging.getLogger(__name__)
class DataValidator:
"""数据验证器
检查数据集的质量、格式和完整性
Schema 契约
~~~~~~~~~~~~
FlorenceForge 定义两种数据 schema:
1. **Training Schema** (``image/prefix/suffix``) — 唯一的训练消费格式。
所有训练器(v1/v2/TVP)均消费此格式。数据转换器(converter_*)
的输出也统一为此格式。
2. **Conversation Schema** (``image/conversations``) — 仅作为数据转换
输入格式。当检测到此格式时,验证器会发出提示,建议通过转换器
转为 Training Schema 后再用于训练。
可通过 ``schema_version`` 参数显式指定期望的 schema,避免自动检测
的歧义。
"""
# Schema 版本标识常量
SCHEMA_TRAINING = "training" # image/prefix/suffix
SCHEMA_CONVERSATION = "conversation" # image/conversations
def __init__(self, strict_mode: bool = False, schema_version: str = ""):
"""初始化验证器
Args:
strict_mode: 是否启用严格模式
schema_version: 期望的 schema 版本 ("training" 或 "conversation")。
为空则自动检测。
"""
self.strict_mode = strict_mode
self.schema_version = schema_version
self.validation_results = []
self.error_count = 0
self.warning_count = 0
self._detected_schema: str = schema_version
def validate_dataset(self, data_path: Union[str, Path]) -> Dict[str, Any]:
"""验证数据集
Args:
data_path: 数据集路径
Returns:
验证结果字典
"""
self.reset()
data_path = Path(data_path)
if not data_path.exists():
self._add_error(f"数据集路径不存在: {data_path}")
return self._get_validation_summary()
# 验证JSON格式
if data_path.suffix == ".json":
return self._validate_json_dataset(data_path)
elif data_path.suffix == ".jsonl":
return self._validate_jsonl_dataset(data_path)
else:
self._add_error(f"不支持的数据格式: {data_path.suffix}")
return self._get_validation_summary()
def _validate_json_dataset(self, data_path: Path) -> Dict[str, Any]:
"""验证JSON格式数据集"""
try:
with open(data_path, "r", encoding="utf-8") as f:
data = json.load(f)
if isinstance(data, list):
return self._validate_sample_list(data, data_path.parent)
elif isinstance(data, dict):
if "data" in data:
return self._validate_sample_list(data["data"], data_path.parent)
else:
self._add_error("JSON数据格式错误:缺少'data'字段")
else:
self._add_error("JSON数据格式错误:根节点必须是列表或字典")
except json.JSONDecodeError as e:
self._add_error(f"JSON解析错误: {e}")
except Exception as e:
self._add_error(f"读取文件错误: {e}")
return self._get_validation_summary()
def _validate_jsonl_dataset(self, data_path: Path) -> Dict[str, Any]:
"""验证JSONL格式数据集"""
samples = []
line_num = 0
try:
with open(data_path, "r", encoding="utf-8") as f:
for line in f:
line_num += 1
line = line.strip()
if not line:
continue
try:
sample = json.loads(line)
samples.append(sample)
except json.JSONDecodeError as e:
self._add_error(f"第{line_num}行JSON解析错误: {e}")
return self._validate_sample_list(samples, data_path.parent)
except Exception as e:
self._add_error(f"读取JSONL文件错误: {e}")
return self._get_validation_summary()
def _validate_sample_list(
self, samples: List[Dict], base_path: Path
) -> Dict[str, Any]:
"""验证样本列表"""
if not samples:
self._add_error("数据集为空")
return self._get_validation_summary()
# 统计信息
task_counts = Counter()
image_formats = Counter()
missing_images = []
invalid_samples = set()
for i, sample in enumerate(samples):
try:
# 验证必需字段
validation_result = self._validate_sample_fields(sample, i)
if not validation_result:
invalid_samples.add(i)
continue
# 统计任务类型
task_type = sample.get("task_type", "unknown")
task_counts[task_type] += 1
# 验证图像文件
image_path = sample.get("image")
if image_path:
full_image_path = base_path / image_path
if not full_image_path.exists():
missing_images.append((i, image_path))
invalid_samples.add(i)
else:
# 检查图像格式
try:
with Image.open(full_image_path) as img:
image_formats[img.format] += 1
# 检查图像尺寸
if img.size[0] < 32 or img.size[1] < 32:
self._add_warning(
f"样本{i}: 图像尺寸过小 {img.size}"
)
except Exception as e:
self._add_error(f"样本{i}: 无法读取图像 {image_path}: {e}")
invalid_samples.add(i)
# 验证任务特定内容
self._validate_task_specific_content(sample, i)
except Exception as e:
self._add_error(f"样本{i}: 验证过程出错: {e}")
invalid_samples.add(i)
# 记录缺失图像
if missing_images:
for sample_idx, image_path in missing_images[:10]: # 只显示前10个
self._add_error(f"样本{sample_idx}: 图像文件不存在: {image_path}")
if len(missing_images) > 10:
self._add_error(f"还有{len(missing_images) - 10}个样本的图像文件缺失")
# 生成验证报告
summary = self._get_validation_summary()
summary.update(
{
"total_samples": len(samples),
"valid_samples": len(samples) - len(invalid_samples),
"invalid_samples": len(invalid_samples),
"missing_images": len(missing_images),
"task_distribution": dict(task_counts),
"image_formats": dict(image_formats),
}
)
return summary
def _validate_sample_fields(self, sample: Dict[str, Any], index: int) -> bool:
"""验证样本字段,支持 training schema (prefix/suffix) 和 conversations schema"""
# 检测 schema 类型
has_training_fields = "prefix" in sample or "suffix" in sample
has_conversations = "conversations" in sample
# 如果已明确指定 training schema,强制使用 training 验证
if self._detected_schema == "training" or has_training_fields:
if not self._detected_schema:
self._detected_schema = "training"
# Training schema: image + prefix + suffix
if "image" not in sample:
self._add_error(f"样本{index}: 缺少必需字段 'image'")
return False
if "prefix" not in sample:
self._add_error(f"样本{index}: 缺少必需字段 'prefix'")
return False
if not sample.get("prefix", "").strip():
self._add_error(f"样本{index}: prefix 字段不能为空")
return False
if "suffix" not in sample:
self._add_error(f"样本{index}: 缺少必需字段 'suffix'")
return False
if not sample.get("suffix", "").strip():
self._add_warning(f"样本{index}: suffix 字段为空")
return True
elif has_conversations or self._detected_schema == "conversations":
# Conversations schema — 仅作为数据转换输入格式
if not self._detected_schema:
self._detected_schema = "conversation"
required_fields = ["image", "conversations"]
# 检查必需字段
for field in required_fields:
if field not in sample:
self._add_error(f"样本{index}: 缺少必需字段 '{field}'")
return False
# 验证conversations格式
conversations = sample.get("conversations", [])
if not isinstance(conversations, list) or len(conversations) == 0:
self._add_error(f"样本{index}: conversations字段必须是非空列表")
return False
# 验证对话格式
for j, conv in enumerate(conversations):
if not isinstance(conv, dict):
self._add_error(f"样本{index}: conversations[{j}]必须是字典")
return False
if "from" not in conv or "value" not in conv:
self._add_error(
f"样本{index}: conversations[{j}]缺少'from'或'value'字段"
)
return False
if conv["from"] not in ["human", "gpt"]:
self._add_warning(
f"样本{index}: conversations[{j}]['from']值异常: {conv['from']}"
)
return True
else:
# 未知 schema
self._add_error(f"样本{index}: 缺少必需字段 'conversations'")
return False
def _validate_task_specific_content(
self, sample: Dict[str, Any], index: int
) -> None:
"""验证任务特定内容"""
task_type = sample.get("task_type", "")
# Training schema: 检查 prefix/suffix 内容
if "prefix" in sample or "suffix" in sample:
prefix = sample.get("prefix", "")
suffix = sample.get("suffix", "")
if task_type and prefix:
# 验证任务前缀是否匹配
from ..core.tasks import FLORENCE2_TASKS
task_config = FLORENCE2_TASKS.get(task_type)
expected_prompt = task_config.prompt if task_config else ""
if expected_prompt and expected_prompt not in prefix:
self._add_warning(
f"样本{index}: 任务类型 '{task_type}' 的前缀不包含预期提示 '{expected_prompt}'"
)
# 检查 suffix 中的坐标是否超出范围
import re
loc_pattern = re.compile(r"<loc_(\d+)>")
for match in loc_pattern.finditer(suffix):
coord_value = int(match.group(1))
if coord_value > 1000:
self._add_warning(
f"样本{index}: 检测坐标超出范围 <loc_{coord_value}>,有效范围 0-1000"
)
return
# Conversations schema
conversations = sample.get("conversations", [])
if not conversations:
return
# 获取最后一个回答(通常是模型的输出)
last_response = None
for conv in reversed(conversations):
if conv.get("from") == "gpt":
last_response = conv.get("value", "")
break
if not last_response:
self._add_warning(f"样本{index}: 没有找到模型回答")
return
# 根据任务类型验证内容格式
if "detection" in task_type.lower() or "object" in task_type.lower():
self._validate_detection_format(last_response, index)
elif "caption" in task_type.lower() or "description" in task_type.lower():
self._validate_caption_format(last_response, index)
elif "ocr" in task_type.lower():
self._validate_ocr_format(last_response, index)
elif "segmentation" in task_type.lower():
self._validate_segmentation_format(last_response, index)
def _validate_detection_format(self, response: str, index: int) -> None:
"""验证检测格式"""
# 检查Florence-2检测格式
florence_pattern = r"\w+<loc_\d+><loc_\d+><loc_\d+><loc_\d+>"
if re.search(florence_pattern, response):
# 验证坐标范围
coord_pattern = r"<loc_(\d+)>"
coords = re.findall(coord_pattern, response)
for coord in coords:
if int(coord) > 1000:
self._add_warning(f"样本{index}: 检测坐标超出范围: {coord}")
else:
# 尝试JSON格式
try:
data = json.loads(response)
if not isinstance(data, (list, dict)):
self._add_warning(f"样本{index}: 检测结果格式可能不正确")
except json.JSONDecodeError:
self._add_warning(f"样本{index}: 检测结果格式无法识别")
def _validate_caption_format(self, response: str, index: int) -> None:
"""验证描述格式"""
if len(response.strip()) < 5:
self._add_warning(f"样本{index}: 图像描述过短")
elif len(response) > 500:
self._add_warning(f"样本{index}: 图像描述过长")
def _validate_ocr_format(self, response: str, index: int) -> None:
"""验证OCR格式"""
# OCR结果通常是纯文本
if not response.strip():
self._add_warning(f"样本{index}: OCR结果为空")
def _validate_segmentation_format(self, response: str, index: int) -> None:
"""验证分割格式"""
# 检查多边形格式
if "<poly>" not in response and "<seg>" not in response:
self._add_warning(f"样本{index}: 分割结果格式可能不正确")
def _add_error(self, message: str) -> None:
"""添加错误信息"""
self.validation_results.append({"type": "error", "message": message})
self.error_count += 1
logger.error(message)
def _add_warning(self, message: str) -> None:
"""添加警告信息"""
self.validation_results.append({"type": "warning", "message": message})
self.warning_count += 1
logger.warning(message)
def _get_validation_summary(self) -> Dict[str, Any]:
"""获取验证摘要"""
is_valid = self.error_count == 0 and (
not self.strict_mode or self.warning_count == 0
)
result = {
"is_valid": is_valid,
"error_count": self.error_count,
"warning_count": self.warning_count,
"validation_results": self.validation_results,
"status": "passed" if is_valid else "failed",
}
if self._detected_schema:
result["effective_schema"] = self._detected_schema
result["detected_schema"] = self._detected_schema
# 对 conversation schema 添加转换提示(不影响 validation_results 列表)
if self._detected_schema == "conversation":
result["schema_note"] = (
"检测到 conversations schema。此格式仅作为数据转换输入格式,"
"训练时需转换为 training schema (image/prefix/suffix)。"
"请使用 data/converter_*.py 转换器进行格式转换。"
)
return result
def reset(self) -> None:
"""重置验证状态"""
self.validation_results = []
self.error_count = 0
self.warning_count = 0
self._detected_schema = ""
@staticmethod
def validate_florence2_jsonl(
jsonl_path: str,
image_base_path: str = "",
) -> Dict[str, Any]:
"""验证 Florence-2 JSONL 格式数据(静态入口,供转换管线使用)。"""
logger.info("验证JSONL数据: %s", jsonl_path)
jsonl_path = Path(jsonl_path).absolute()
image_base_path = (
Path(image_base_path).absolute() if image_base_path else Path()
)
report: Dict[str, Any] = {
"total_samples": 0,
"valid_samples": 0,
"invalid_samples": 0,
"missing_images": 0,
"task_distribution": {},
"errors": [],
}
with open(jsonl_path, "r", encoding="utf-8") as handle:
for line_num, line in enumerate(tqdm(handle, desc="验证JSONL进度"), 1):
report["total_samples"] += 1
try:
data = json.loads(line.strip())
for field in ("image", "prefix", "suffix"):
if field not in data:
report["errors"].append(
f"第{line_num}行: 缺少字段 '{field}'"
)
continue
image_path = Path(data["image"])
if not image_path.exists():
report["missing_images"] += 1
report["errors"].append(
f"第{line_num}行: 图像文件不存在 '{image_path}'"
)
for field in ("label_file", "txt_file", "xml_file", "mask_dir"):
if field in data and not Path(data[field]).exists():
report["errors"].append(
f"第{line_num}行: 标签文件不存在 '{data[field]}'"
)
task_type = data["prefix"].strip("<>")
report["task_distribution"][task_type] = (
report["task_distribution"].get(task_type, 0) + 1
)
report["valid_samples"] += 1
except json.JSONDecodeError as exc:
report["invalid_samples"] += 1
report["errors"].append(f"第{line_num}行: JSON解析错误 - {exc}")
except Exception as exc:
report["invalid_samples"] += 1
report["errors"].append(f"第{line_num}行: 验证错误 - {exc}")
logger.info(
"验证完成: 总计%s样本,有效%s,无效%s",
report["total_samples"],
report["valid_samples"],
report["invalid_samples"],
)
return report
@staticmethod
def generate_validation_report(report: Dict[str, Any], output_path: str) -> None:
"""将 ``validate_florence2_jsonl`` 结果写入 Markdown 报告。"""
output_path = Path(output_path).absolute()
output_path.parent.mkdir(parents=True, exist_ok=True)
with open(output_path, "w", encoding="utf-8") as handle:
handle.write("# 数据验证报告\n\n")
handle.write(f"- **总样本数**: {report['total_samples']}\n")
handle.write(f"- **有效样本数**: {report['valid_samples']}\n")
handle.write(f"- **无效样本数**: {report['invalid_samples']}\n")
handle.write(f"- **缺失图像数**: {report['missing_images']}\n\n")
handle.write("## 任务分布\n\n")
for task, count in report["task_distribution"].items():
handle.write(f"- **{task}**: {count} 样本\n")
if report["errors"]:
handle.write("\n## 错误详情\n\n")
for error in report["errors"][:100]:
handle.write(f"- {error}\n")
if len(report["errors"]) > 100:
handle.write(
f"\n... 还有 {len(report['errors']) - 100} 个错误未显示\n"
)
def validate_data_format(
data_path: Union[str, Path], strict_mode: bool = False, schema: str = None
) -> Dict[str, Any]:
"""验证数据格式的便捷函数
Args:
data_path: 数据路径
strict_mode: 是否启用严格模式
schema: 可选,指定数据 schema("training" 或 "conversations")
Returns:
验证结果
"""
validator = DataValidator(strict_mode=strict_mode)
if schema:
validator._detected_schema = schema
return validator.validate_dataset(data_path)
|