fix

api_routes.py

@@ -731,7 +731,7 @@ async def _refresh_celebrity_cache(sample_image_path: str,
                 img_path=sample_image_path,
                 db_path=db_path,
                 model_name="ArcFace",
-                detector_backend="retinaface",
+                detector_backend="yolov11n",
                 distance_metric="cosine",
                 enforce_detection=True,
                 silent=True,
@@ -748,7 +748,7 @@ async def _refresh_celebrity_cache(sample_image_path: str,
                         img_path=sample_image_path,
                         db_path=db_path,
                         model_name="ArcFace",
-                        detector_backend="retinaface",
+                        detector_backend="yolov11n",
                         distance_metric="cosine",
                         enforce_detection=True,
                         silent=True,
@@ -768,7 +768,7 @@ async def _refresh_celebrity_cache(sample_image_path: str,
                     img_path=sample_image_path,
                     db_path=db_path,
                     model_name="ArcFace",
-                    detector_backend="retinaface",
+                    detector_backend="yolov11n",
                     distance_metric="cosine",
                     enforce_detection=True,
                     silent=True,
@@ -3905,7 +3905,7 @@ async def match_celebrity_face(
                 img_path=temp_path,
                 db_path=db_path,
                 model_name="ArcFace",
-                detector_backend="retinaface",
+                detector_backend="yolov11n",
                 distance_metric="cosine",
                 enforce_detection=True,
                 silent=True,
@@ -4224,7 +4224,7 @@ async def face_similarity_verification(
                     img1_path=original_path1,
                     img2_path=original_path2,
                     model_name="ArcFace",
-                    detector_backend="retinaface",
+                    detector_backend="yolov11n",
                     distance_metric="cosine"
                 )
                 logger.info(
@@ -4240,7 +4240,7 @@ async def face_similarity_verification(
                             img1_path=original_path1,
                             img2_path=original_path2,
                             model_name="ArcFace",
-                            detector_backend="retinaface",
+                            detector_backend="yolov11n",
                             distance_metric="cosine"
                         )
                         logger.info(
@@ -4262,7 +4262,7 @@ async def face_similarity_verification(
                         img1_path=original_path1,
                         img2_path=original_path2,
                         model_name="ArcFace",
-                        detector_backend="retinaface",
+                        detector_backend="yolov11n",
                         distance_metric="cosine"
                     )
                     logger.info(

test/celebrity_crawler.py

@@ -0,0 +1,227 @@
+import time
+from io import BytesIO
+from pathlib import Path
+
+import requests
+from PIL import Image
+
+
+class CelebrityCrawler:
+  def __init__(self, output_dir="celebrity_images"):
+    self.output_dir = output_dir
+    self.headers = {
+      'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36'
+    }
+    Path(output_dir).mkdir(parents=True, exist_ok=True)
+
+  def read_celebrities_from_txt(self, file_path):
+    """
+    从txt文件读取明星信息
+    支持格式：
+    1. 姓名,职业
+    2. 姓名
+    """
+    celebrities = []
+    with open(file_path, 'r', encoding='utf-8') as f:
+      for line in f:
+        line = line.strip()
+        if not line or line.startswith('#'):
+          continue
+
+        parts = line.split(',')
+        name = parts[0].strip()
+        profession = parts[1].strip() if len(parts) > 1 else "明星"
+
+        celebrities.append({
+          'name': name,
+          'profession': profession
+        })
+    return celebrities
+
+  def search_bing_images(self, celebrity_name, max_images=20):
+    """使用Bing图片搜索API获取图片URL"""
+    search_url = "https://www.bing.com/images/search"
+    params = {
+      'q': celebrity_name + " 明星",
+      'first': 0,
+      'count': max_images
+    }
+
+    try:
+      response = requests.get(search_url, params=params, headers=self.headers,
+                              timeout=10)
+      response.raise_for_status()
+
+      # 简单的HTML解析获取图片URL
+      import re
+      img_urls = re.findall(r'murl":"(.*?)"', response.text)
+      return img_urls[:max_images]
+    except Exception as e:
+      print(f"搜索 {celebrity_name} 时出错: {e}")
+      return []
+
+  def search_baidu_images(self, celebrity_name, max_images=20):
+    """使用百度图片搜索获取图片URL"""
+    search_url = "https://image.baidu.com/search/acjson"
+    params = {
+      'tn': 'resultjson_com',
+      'word': celebrity_name + " 明星",
+      'pn': 0,
+      'rn': max_images,
+      'ie': 'utf-8'
+    }
+
+    try:
+      response = requests.get(search_url, params=params, headers=self.headers,
+                              timeout=10)
+      response.raise_for_status()
+      data = response.json()
+
+      img_urls = []
+      if 'data' in data:
+        for item in data['data']:
+          if 'thumbURL' in item:
+            img_urls.append(item['thumbURL'])
+      return img_urls[:max_images]
+    except Exception as e:
+      print(f"搜索 {celebrity_name} 时出错: {e}")
+      return []
+
+  def download_image(self, url, save_path):
+    """下载单张图片"""
+    try:
+      response = requests.get(url, headers=self.headers, timeout=15)
+      response.raise_for_status()
+
+      # 验证是否为有效图片
+      img = Image.open(BytesIO(response.content))
+
+      # 过滤太小的图片
+      if img.size[0] < 100 or img.size[1] < 100:
+        return False
+
+      # 保存图片
+      img = img.convert('RGB')
+      img.save(save_path, 'JPEG', quality=95)
+      return True
+    except Exception as e:
+      print(f"  下载失败: {str(e)[:50]}")
+      return False
+
+  def crawl_celebrity_images(self, celebrity, max_images=20,
+      search_engine='baidu'):
+    """爬取单个明星的图片"""
+    name = celebrity['name']
+    print(f"\n正在爬取: {name} ({celebrity['profession']})")
+
+    # 创建明星专属文件夹
+    celebrity_dir = Path(self.output_dir)
+    celebrity_dir.mkdir(parents=True, exist_ok=True)
+
+    # 获取图片URL列表
+    if search_engine == 'baidu':
+      img_urls = self.search_baidu_images(name, max_images * 2)
+    else:
+      img_urls = self.search_bing_images(name, max_images * 2)
+
+    if not img_urls:
+      print(f"  未找到 {name} 的图片")
+      return 0
+
+    print(f"  找到 {len(img_urls)} 个图片链接")
+
+    # 下载图片
+    success_count = 0
+    for idx, url in enumerate(img_urls):
+      if success_count >= max_images:
+        break
+
+      save_path = celebrity_dir / f"{name}_{idx + 1:03d}.jpg"
+
+      # 跳过已存在的文件
+      if save_path.exists():
+        success_count += 1
+        continue
+
+      print(f"  下载 {idx + 1}/{len(img_urls)}...", end=' ')
+      if self.download_image(url, save_path):
+        success_count += 1
+        print("✓")
+      else:
+        print("✗")
+
+      # 避免请求过快
+      time.sleep(0.5)
+
+    print(f"  成功下载 {success_count} 张图片")
+    return success_count
+
+  def crawl_all(self, txt_file, max_images_per_celebrity=20,
+      search_engine='baidu'):
+    """爬取所有明星的图片"""
+    print("=" * 60)
+    print("明星照片爬取工具")
+    print("=" * 60)
+
+    # 读取明星列表
+    celebrities = self.read_celebrities_from_txt(txt_file)
+    print(f"\n从 {txt_file} 读取到 {len(celebrities)} 位明星")
+
+    # 统计信息
+    total_images = 0
+    failed_celebrities = []
+
+    # 爬取每位明星
+    for i, celebrity in enumerate(celebrities, 1):
+      print(f"\n[{i}/{len(celebrities)}]", end=' ')
+
+      try:
+        count = self.crawl_celebrity_images(
+            celebrity,
+            max_images=max_images_per_celebrity,
+            search_engine=search_engine
+        )
+        total_images += count
+
+        if count == 0:
+          failed_celebrities.append(celebrity['name'])
+
+        # 每爬取5个明星后暂停一下
+        if i % 5 == 0:
+          print(f"\n  已完成 {i}/{len(celebrities)}, 休息3秒...")
+          time.sleep(3)
+
+      except Exception as e:
+        print(f"  处理 {celebrity['name']} 时出错: {e}")
+        failed_celebrities.append(celebrity['name'])
+
+    # 输出统计
+    print("\n" + "=" * 60)
+    print("爬取完成！")
+    print("=" * 60)
+    print(f"总明星数: {len(celebrities)}")
+    print(f"成功爬取: {len(celebrities) - len(failed_celebrities)}")
+    print(f"失败数量: {len(failed_celebrities)}")
+    print(f"总图片数: {total_images}")
+    print(f"保存位置: {self.output_dir}")
+
+    if failed_celebrities:
+      print(f"\n失败的明星: {', '.join(failed_celebrities)}")
+
+
+# 使用示例
+if __name__ == "__main__":
+  # 创建爬虫实例
+  crawler = CelebrityCrawler(output_dir="celebrity_dataset")
+
+  # 从txt文件爬取
+  # txt文件格式示例：
+  # 周杰伦,歌手
+  # 刘德华,演员
+  # 范冰冰,演员
+
+  crawler.crawl_all(
+      txt_file="celebrity_real_names.txt",  # 你的txt文件路径
+      max_images_per_celebrity=1,  # 每位明星爬取的图片数量
+      search_engine='baidu'  # 'baidu' 或 'bing'
+  )

test/dow_img.py

@@ -0,0 +1,24 @@
+import cv2
+
+# 读取图片
+img = cv2.imread("~/Pictures/header.png")
+
+# 设置压缩质量（0-100，值越小压缩越狠，质量越差）
+quality = 50
+
+# 写入压缩后的图像（注意必须是 .webp）
+cv2.imwrite(
+    "~/Pictures/output_small.webp",
+    img,
+    [int(cv2.IMWRITE_WEBP_QUALITY), quality],
+)
+
+
+# # 读取原图
+# img = cv2.imread("~/Pictures/header.png")
+#
+# # 缩放图像（例如缩小为原图的一半）
+# resized = cv2.resize(img, (img.shape[1] // 2, img.shape[0] // 2))
+#
+# # 写入压缩图像，降低质量
+# cv2.imwrite("~/Pictures/output_small.webp", resized, [int(cv2.IMWRITE_WEBP_QUALITY), 40])

test/howcuteami.py

@@ -0,0 +1,202 @@
+import cv2
+import math
+import argparse
+import numpy as np
+import os
+
+
+# detect face
+def highlightFace(net, frame, conf_threshold=0.95):
+    frameOpencvDnn = frame.copy()
+    frameHeight = frameOpencvDnn.shape[0]
+    frameWidth = frameOpencvDnn.shape[1]
+    blob = cv2.dnn.blobFromImage(
+        frameOpencvDnn, 1.0, (300, 300), [104, 117, 123], True, False
+    )
+
+    net.setInput(blob)
+    detections = net.forward()
+    faceBoxes = []
+
+    for i in range(detections.shape[2]):
+        confidence = detections[0, 0, i, 2]
+        if confidence > conf_threshold:
+            x1 = int(detections[0, 0, i, 3] * frameWidth)
+            y1 = int(detections[0, 0, i, 4] * frameHeight)
+            x2 = int(detections[0, 0, i, 5] * frameWidth)
+            y2 = int(detections[0, 0, i, 6] * frameHeight)
+            faceBoxes.append(scale([x1, y1, x2, y2]))
+
+    return faceBoxes
+
+
+# scale current rectangle to box
+def scale(box):
+    width = box[2] - box[0]
+    height = box[3] - box[1]
+    maximum = max(width, height)
+    dx = int((maximum - width) / 2)
+    dy = int((maximum - height) / 2)
+
+    bboxes = [box[0] - dx, box[1] - dy, box[2] + dx, box[3] + dy]
+    return bboxes
+
+
+# crop image
+def cropImage(image, box):
+    num = image[box[1] : box[3], box[0] : box[2]]
+    return num
+
+
+# main
+parser = argparse.ArgumentParser()
+parser.add_argument("-i", "--image", type=str, required=False, help="input image")
+args = parser.parse_args()
+
+# 创建输出目录
+output_dir = "../output"
+if not os.path.exists(output_dir):
+    os.makedirs(output_dir)
+
+faceProto = "models/opencv_face_detector.pbtxt"
+faceModel = "models/opencv_face_detector_uint8.pb"
+ageProto = "models/age_googlenet.prototxt"
+ageModel = "models/age_googlenet.caffemodel"
+genderProto = "models/gender_googlenet.prototxt"
+genderModel = "models/gender_googlenet.caffemodel"
+beautyProto = "models/beauty_resnet.prototxt"
+beautyModel = "models/beauty_resnet.caffemodel"
+
+MODEL_MEAN_VALUES = (104, 117, 123)
+ageList = [
+    "(0-2)",
+    "(4-6)",
+    "(8-12)",
+    "(15-20)",
+    "(25-32)",
+    "(38-43)",
+    "(48-53)",
+    "(60-100)",
+]
+genderList = ["Male", "Female"]
+
+# 定义性别对应的颜色 (BGR格式)
+gender_colors = {
+    "Male": (255, 165, 0),  # 橙色 Orange
+    "Female": (255, 0, 255),  # 洋红 Magenta / Fuchsia
+}
+
+faceNet = cv2.dnn.readNet(faceModel, faceProto)
+ageNet = cv2.dnn.readNet(ageModel, ageProto)
+genderNet = cv2.dnn.readNet(genderModel, genderProto)
+beautyNet = cv2.dnn.readNet(beautyModel, beautyProto)
+
+# 读取图片
+image_path = args.image if args.image else "images/charlize.jpg"
+frame = cv2.imread(image_path)
+
+if frame is None:
+    print(f"无法读取图片: {image_path}")
+    exit()
+
+faceBoxes = highlightFace(faceNet, frame)
+if not faceBoxes:
+    print("No face detected")
+    exit()
+
+print(f"检测到 {len(faceBoxes)} 张人脸")
+
+for i, faceBox in enumerate(faceBoxes):
+    # 提取人脸区域
+    face = cropImage(frame, faceBox)
+    face_resized = cv2.resize(face, (224, 224))
+
+    # gender net
+    blob = cv2.dnn.blobFromImage(
+        face_resized, 1.0, (224, 224), MODEL_MEAN_VALUES, swapRB=False
+    )
+    genderNet.setInput(blob)
+    genderPreds = genderNet.forward()
+    gender = genderList[genderPreds[0].argmax()]
+    print(f"Gender: {gender}")
+
+    # age net
+    ageNet.setInput(blob)
+    agePreds = ageNet.forward()
+    age = ageList[agePreds[0].argmax()]
+    print(f"Age: {age[1:-1]} years")
+
+    # beauty net
+    blob = cv2.dnn.blobFromImage(
+        face_resized, 1.0 / 255, (224, 224), MODEL_MEAN_VALUES, swapRB=False
+    )
+    beautyNet.setInput(blob)
+    beautyPreds = beautyNet.forward()
+    beauty = round(2.0 * sum(beautyPreds[0]), 1)
+    print(f"Beauty: {beauty}/10.0")
+
+    # 根据性别选择颜色
+    color = gender_colors[gender]
+
+    # 保存人脸图片 - 使用cv2.imwrite
+    face_filename = f"{output_dir}/face_{i+1}.webp"
+    cv2.imwrite(face_filename, face, [cv2.IMWRITE_WEBP_QUALITY, 95])
+    print(f"人脸图片已保存: {face_filename}")
+
+    # 保存评分到图片上（可选）
+    face_with_text = face.copy()
+    cv2.putText(
+        face_with_text, f"{gender}", (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.7, color, 2
+    )
+    cv2.putText(
+        face_with_text,
+        f"{age[1:-1]} years",
+        (10, 60),
+        cv2.FONT_HERSHEY_SIMPLEX,
+        0.7,
+        color,
+        2,
+    )
+    cv2.putText(
+        face_with_text,
+        f"{beauty}/10.0",
+        (10, 90),
+        cv2.FONT_HERSHEY_SIMPLEX,
+        0.7,
+        color,
+        2,
+    )
+
+    annotated_filename = f"{output_dir}/face_{i+1}_annotated.webp"
+    cv2.imwrite(annotated_filename, face_with_text, [cv2.IMWRITE_WEBP_QUALITY, 95])
+    print(f"标注人脸已保存: {annotated_filename}")
+
+    # 在原图上绘制人脸框和信息
+    cv2.rectangle(
+        frame,
+        (faceBox[0], faceBox[1]),
+        (faceBox[2], faceBox[3]),
+        color,
+        int(round(frame.shape[0] / 400)),
+        8,
+    )
+    cv2.putText(
+        frame,
+        f"{gender}, {age}, {beauty}",
+        (faceBox[0], faceBox[1] - 10),
+        cv2.FONT_HERSHEY_SIMPLEX,
+        1.25,
+        color,
+        2,
+        cv2.LINE_AA,
+    )
+
+# 保存完整的标注图片
+result_filename = f"{output_dir}/result_full.webp"
+cv2.imwrite(result_filename, frame, [cv2.IMWRITE_WEBP_QUALITY, 95])
+print(f"完整结果图片已保存: {result_filename}")
+
+# 显示图片
+cv2.imshow("howbeautifulami", frame)
+cv2.waitKey(0)
+cv2.destroyAllWindows()

test/import_history_images.py

@@ -0,0 +1,162 @@
+#!/usr/bin/env python3
+"""
+导入历史图片文件到数据库的脚本
+"""
+
+import asyncio
+import hashlib
+import os
+import sys
+import time
+from datetime import datetime
+from pathlib import Path
+
+# 添加项目根目录到Python路径
+sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+from database import record_image_creation, fetch_records_by_paths
+
+
+def calculate_file_hash(file_path):
+  """计算文件的MD5哈希值"""
+  hash_md5 = hashlib.md5()
+  with open(file_path, "rb") as f:
+    # 分块读取文件，避免大文件占用过多内存
+    for chunk in iter(lambda: f.read(4096), b""):
+      hash_md5.update(chunk)
+  return hash_md5.hexdigest()
+
+
+def infer_category_from_filename(filename):
+  """从文件名推断类别"""
+  filename_lower = filename.lower()
+
+  # 处理动漫风格化类型
+  if '_anime_style_' in filename_lower:
+    return 'anime_style'
+
+  # 查找最后一个下划线和第一个点的位置
+  last_underscore_index = filename_lower.rfind('_')
+  first_dot_index = filename_lower.find('.', last_underscore_index)
+
+  # 如果找到了下划线和点，且下划线在点之前
+  if last_underscore_index != -1 and first_dot_index != -1 and last_underscore_index < first_dot_index:
+    # 提取下划线和点之间的内容
+    file_type = filename_lower[last_underscore_index + 1:first_dot_index]
+
+    # 根据类型返回中文描述
+    type_mapping = {
+      'restore': 'restore',
+      'upcolor': 'upcolor',
+      'grayscale': 'grayscale',
+      'upscale': 'upscale',
+      'compress': 'compress',
+      'id_photo': 'id_photo',
+      'grid': 'grid',
+      'rvm': 'rvm',
+      'celebrity': 'celebrity',
+      'face': 'face',
+      'original': 'original'
+    }
+
+    return type_mapping.get(file_type, 'other')
+
+  # 默认返回 other
+  return 'other'
+
+
+async def import_history_images(source_dir, nickname="system_import"):
+  """导入历史图片到数据库"""
+  source_path = Path(source_dir)
+
+  if not source_path.exists():
+    print(f"错误: 目录 {source_dir} 不存在")
+    return
+
+  # 支持的图片格式
+  image_extensions = {'.jpg', '.jpeg', '.png', '.webp', '.bmp', '.gif', '.tiff',
+                      '.tif'}
+
+  # 获取所有图片文件
+  image_files = []
+  for ext in image_extensions:
+    image_files.extend(source_path.glob(f"*{ext}"))
+    image_files.extend(source_path.glob(f"*{ext.upper()}"))
+
+  print(f"找到 {len(image_files)} 个图片文件")
+
+  imported_count = 0
+  skipped_count = 0
+
+  for image_path in image_files:
+    try:
+      file_name = image_path.name
+
+      # 检查文件是否已存在于数据库中（基于文件名）
+      records = await fetch_records_by_paths([file_name])
+
+      if file_name in records:
+        print(f"跳过已存在的文件: {file_name}")
+        skipped_count += 1
+        continue
+
+      # 如果数据库中没有记录，则继续导入
+      # 计算文件哈希值用于进一步确认唯一性
+      file_hash = calculate_file_hash(str(image_path))
+
+      # 推断文件类别
+      category = infer_category_from_filename(file_name)
+
+      # 记录到数据库
+      await record_image_creation(
+          file_path=file_name,  # 使用文件名而不是完整路径
+          nickname=nickname,
+          category=category,
+          bos_uploaded=False,  # 历史文件通常未上传到BOS
+          score=0.0,  # 历史文件默认分数为0
+          extra_metadata={
+            "source": "history_import",
+            "original_path": str(image_path),
+            "file_hash": file_hash,
+            "import_time": datetime.now().isoformat()
+          }
+      )
+
+      imported_count += 1
+      print(f"成功导入: {file_name} (类别: {category})")
+
+    except Exception as e:
+      print(f"导入文件失败 {image_path.name}: {str(e)}")
+      continue
+
+  print(f"\n导入完成!")
+  print(f"成功导入: {imported_count} 个文件")
+  print(f"跳过: {skipped_count} 个文件")
+
+
+async def main():
+  if len(sys.argv) < 2:
+    print("用法: python import_history_images.py <图片目录路径> [昵称]")
+    print(
+      "示例: python import_history_images.py /Users/chenchaoyun/app/data/images")
+    print(
+      "示例: python import_history_images.py /Users/chenchaoyun/app/data/images \"历史导入\"")
+    sys.exit(1)
+
+  source_directory = sys.argv[1]
+  nickname = sys.argv[2] if len(sys.argv) > 2 else "system_import"
+
+  print(f"开始导入图片文件...")
+  print(f"源目录: {source_directory}")
+  print(f"用户昵称: {nickname}")
+  print("-" * 50)
+
+  start_time = time.time()
+  await import_history_images(source_directory, nickname)
+  end_time = time.time()
+
+  print(f"\n总耗时: {end_time - start_time:.2f} 秒")
+
+
+if __name__ == "__main__":
+  asyncio.run(main())

test/test_deepface.py

@@ -0,0 +1,38 @@
+import json
+import time
+from deepface import DeepFace
+
+images_path = "/Users/chenchaoyun/Pictures/face"
+
+# ========== 2. 人脸相似度比对 ==========
+start_time = time.time()
+result_verification = DeepFace.verify(
+    img1_path=images_path + "/4.webp",
+    img2_path=images_path + "/5.webp",
+    model_name="ArcFace",                # 指定模型
+    detector_backend="yolov11n",       # 人脸检测器 retinaface / yolov8 / opencv / ssd / mediapipe
+    distance_metric="cosine"             # 相似度度量
+)
+end_time = time.time()
+print(f"🕒 人脸比对耗时: {end_time - start_time:.3f} 秒")
+
+# 打印结果
+print(json.dumps(result_verification, ensure_ascii=False, indent=2))
+
+
+# ========== 1. 人脸识别 ==========
+
+start_time = time.time()
+result_recognition = DeepFace.find(
+    img_path=images_path + "/1.jpg",  # 待识别人脸
+    db_path=images_path,                 # 数据库路径
+    model_name="ArcFace",                # 指定模型
+    detector_backend="yolov11n",       # 人脸检测器
+    distance_metric="cosine"             # 相似度度量
+)
+end_time = time.time()
+print(f"🕒 人脸识别耗时: {end_time - start_time:.3f} 秒")
+
+# 如果需要打印结果，可以取消注释
+# df = result_recognition[0]
+# print(df.to_json(orient="records", force_ascii=False))

test/test_main.http

@@ -0,0 +1,11 @@
+# Test your FastAPI endpoints
+
+GET http://127.0.0.1:8000/
+Accept: application/json
+
+###
+
+GET http://127.0.0.1:8000/hello/User
+Accept: application/json
+
+###

test/test_rvm_infer.py

@@ -0,0 +1,46 @@
+import time
+
+import cv2
+import numpy as np
+import torch
+from torchvision import transforms
+
+device = "cpu"
+
+# 输入输出路径
+input_path = "~/Pictures/face/yang.webp"
+output_path = "~/Pictures/face/output_alpha.webp"
+
+# ✅ 加载预训练模型 (resnet50)
+model = torch.hub.load("PeterL1n/RobustVideoMatting", "resnet50").to(device).eval()
+
+# 开始计时
+start = time.time()
+
+# 读图 (BGR->RGB)
+img = cv2.imread(input_path)[:, :, ::-1].copy()
+src = transforms.ToTensor()(img).unsqueeze(0).to(device)
+
+# 推理
+rec = [None] * 4
+with torch.no_grad():
+  fgr, pha, *rec = model(src, *rec, downsample_ratio=0.25)
+
+# 转 numpy
+fgr = (fgr[0].permute(1, 2, 0).cpu().numpy() * 255).astype(np.uint8)  # (H,W,3)
+pha = (pha[0, 0].cpu().numpy() * 255).astype(np.uint8)                # (H,W)
+
+# 拼接 RGBA
+rgba = np.dstack((fgr, pha))  # (H,W,4)
+
+# 保存 WebP (带透明度)
+cv2.imwrite(output_path, rgba[:, :, [2,1,0,3]], [cv2.IMWRITE_WEBP_QUALITY, 100])  # 转成 BGRA 顺序
+
+# 结束计时
+elapsed = time.time() - start
+
+# 控制台日志输出
+print(f"✅ RVM 抠图完成 (透明背景)")
+print(f"   输入文件: {input_path}")
+print(f"   输出文件: {output_path}")
+print(f"   耗时: {elapsed:.3f} 秒 (设备: {device})")

test/test_score.py

@@ -0,0 +1,26 @@
+import json
+import logging
+
+import numpy as np
+from retinaface import RetinaFace
+
+
+def default_converter(o):
+    if isinstance(o, np.integer):
+        return int(o)
+    if isinstance(o, np.floating):
+        return float(o)
+    if isinstance(o, np.ndarray):
+        return o.tolist()
+    return str(o)
+
+
+# 配置日志
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger(__name__)
+
+resp = RetinaFace.detect_faces("~/Downloads/chounan.jpeg")
+
+logger.info(
+    "search results: " + json.dumps(resp, ensure_ascii=False, default=default_converter)
+)

test/test_score_adjustment_demo.py

@@ -0,0 +1,30 @@
+def adjust_score(score, threshold, gamma):
+    """根据阈值和gamma值调整评分"""
+    if score < threshold:
+        adjusted = threshold - gamma * (threshold - score)
+        return round(min(10.0, max(0.0, adjusted)), 1)
+    return score
+
+# 默认参数 (T=9.0, γ=0.5)
+default_threshold = 9.0
+default_gamma = 0.5
+
+# 新参数1 (T=8.0, γ=0.5)
+new_threshold_1 = 9
+new_gamma_1 = 0.9
+
+# 新参数2 (T=8.0, γ=0.3)
+new_threshold_2 = 9
+new_gamma_2 = 0.8
+
+print(f"原始分\tT={default_threshold},y={default_gamma}\tT={new_threshold_1},γ={new_gamma_1}\tT={new_threshold_2},γ={new_gamma_2}")
+print("-----\t----------\t----------\t----------")
+
+# 从1.0到10.0，以0.1为步长
+for i in range(10, 101):
+    score = i / 10.0
+    default_adjusted = adjust_score(score, default_threshold, default_gamma)
+    new_adjusted_1 = adjust_score(score, new_threshold_1, new_gamma_1)
+    new_adjusted_2 = adjust_score(score, new_threshold_2, new_gamma_2)
+    # 确保显示小数点
+    print(f"{score:.1f}\t\t\t{default_adjusted:.1f}\t\t\t\t\t{new_adjusted_1:.1f}\t\t\t\t\t{new_adjusted_2:.1f}")

test/test_sky.py

@@ -0,0 +1,15 @@
+import os.path as osp
+
+import cv2
+from modelscope.outputs import OutputKeys
+from modelscope.pipelines import pipeline
+from modelscope.utils.constant import Tasks
+
+image_skychange = pipeline(Tasks.image_skychange,
+                           model='iic/cv_hrnetocr_skychange')
+result = image_skychange(
+    {'sky_image': '/Users/chenchaoyun/Downloads/sky_image.jpg',
+     'scene_image': '/Users/chenchaoyun/Pictures/face/NXEo0zusSaNB2fa232c84898e92ff165e2dfee59cb54.jpg'})
+cv2.imwrite('/Users/chenchaoyun/Downloads/result.png',
+            result[OutputKeys.OUTPUT_IMG])
+print(f'Output written to {osp.abspath("result.png")}')