\u60f3\u8c61\u4e00\u4e0b，\u4f60\u624b\u5934\u6709\u4e00\u53f0\u95f2\u7f6e\u7684\u65e7\u670d\u52a1\u5668，\u6216\u8005\u4e00\u4e2a\u914d\u7f6e\u4e0d\u9ad8\u7684\u4e91\u4e3b\u673a，\u53ea\u67092\u4e2aCPU\u6838\u5fc3\u548c2GB\u5185\u5b58\u3002\u4f60\u60f3\u5728\u4e0a\u9762\u90e8\u7f72\u4e00\u4e2a\u5148\u8fdb\u7684\u4eba\u8138\u68c0\u6d4b\u670d\u52a1，\u7528\u6765\u5904\u7406\u56fe\u7247\u3001\u8bc6\u522b\u7528\u6237，\u751a\u81f3\u505a\u4e00\u4e9b\u7b80\u5355\u7684\u5b89\u9632\u76d1\u63a7\u3002\u4f46\u4e00\u770b\u5230\u90a3\u4e9b\u52a8\u8f84\u8981\u6c428GB\u300116GB\u5185\u5b58\u7684AI\u6a21\u578b，\u662f\u4e0d\u662f\u89c9\u5f97\u5e0c\u671b\u6e3a\u832b？<\/p>\n
\u522b\u6025\u7740\u653e\u5f03\u3002\u4eca\u5929\u6211\u8981\u5206\u4eab\u7684，\u5c31\u662f\u5982\u4f55\u5728\u8fd9\u6837\u7684\u201c\u4f4e\u914d\u201d\u73af\u5883\u4e0b，\u6210\u529f\u8fd0\u884cMogFace\u4eba\u8138\u68c0\u6d4b\u6a21\u578b\u53ca\u5176WebUI\u754c\u9762\u3002MogFace\u662fCVPR 2022\u4e0a\u63d0\u51fa\u7684\u4e00\u4e2a\u4f18\u79c0\u4eba\u8138\u68c0\u6d4b\u5668，\u7cbe\u5ea6\u9ad8\u3001\u7a33\u5b9a\u6027\u597d，\u80fd\u8bc6\u522b\u4fa7\u8138\u3001\u6234\u53e3\u7f69\u3001\u5149\u7ebf\u6697\u7b49\u5404\u79cd\u590d\u6742\u573a\u666f\u4e0b\u7684\u4eba\u8138\u3002\u66f4\u91cd\u8981\u7684\u662f，\u7ecf\u8fc7\u4e00\u4e9b\u4f18\u5316\u6280\u5de7，\u5b83\u5b8c\u5168\u53ef\u4ee5\u5728\u8d44\u6e90\u6709\u9650\u7684\u670d\u52a1\u5668\u4e0a\u6d41\u7545\u8fd0\u884c\u3002<\/p>\n
\u8fd9\u7bc7\u6587\u7ae0\u5c31\u662f\u4e3a\u4f60\u51c6\u5907\u7684\u201c\u7626\u8eab\u201d\u6307\u5357\u3002\u6211\u4f1a\u5e26\u4f60\u4e00\u6b65\u6b65\u8c03\u6574\u914d\u7f6e\u3001\u4f18\u5316\u53c2\u6570，\u8ba9\u8fd9\u4e2a\u5f3a\u5927\u7684\u6a21\u578b\u5728\u4f60\u7684\u5c0f\u670d\u52a1\u5668\u4e0a\u201c\u5b89\u5bb6\u843d\u6237\u201d\u3002\u65e0\u8bba\u4f60\u662f\u4e2a\u4eba\u5f00\u53d1\u8005\u3001\u5c0f\u56e2\u961f，\u8fd8\u662f\u53ea\u662f\u60f3\u4f4e\u6210\u672c\u4f53\u9a8cAI\u80fd\u529b，\u8fd9\u5957\u65b9\u6cd5\u90fd\u80fd\u5e2e\u4f60\u7701\u4e0b\u4e0d\u5c11\u786c\u4ef6\u6210\u672c\u3002<\/p>\n
\u5728\u5f00\u59cb\u52a8\u624b\u4e4b\u524d，\u6211\u4eec\u5148\u7b80\u5355\u4e86\u89e3\u4e00\u4e0bMogFace\u6a21\u578b\u7684\u7279\u70b9，\u8fd9\u80fd\u5e2e\u52a9\u6211\u4eec\u66f4\u597d\u5730\u8fdb\u884c\u4f18\u5316\u3002<\/p>\n
MogFace\u5e76\u4e0d\u662f\u4e00\u4e2a\u201c\u8f7b\u91cf\u7ea7\u201d\u6a21\u578b，\u5b83\u7684\u57fa\u7840\u7248\u672c\u57fa\u4e8eResNet101，\u53c2\u6570\u91cf\u4e0d\u5c0f\u3002\u4f46\u5b83\u5728\u8bbe\u8ba1\u4e0a\u6709\u51e0\u4e2a\u7279\u70b9，\u8ba9\u6211\u4eec\u6709\u673a\u4f1a\u5728\u4f4e\u914d\u73af\u5883\u4e0b\u8fd0\u884c：<\/p>\n
\u6211\u4eec\u7684\u76ee\u6807\u73af\u5883\u662f2\u6838CPU\u548c2GB\u5185\u5b58\u3002\u8fd9\u542c\u8d77\u6765\u5f88\u7d27\u5f20，\u4f46\u5e76\u975e\u4e0d\u53ef\u80fd\u3002<\/p>\n
\u4e3b\u8981\u6311\u6218：<\/p>\n
\u6211\u4eec\u7684\u4f18\u5316\u601d\u8def：<\/p>\n
\u7406\u89e3\u4e86\u8fd9\u4e9b，\u6211\u4eec\u5c31\u53ef\u4ee5\u5f00\u59cb\u52a8\u624b\u4e86\u3002<\/p>\n
\u5047\u8bbe\u4f60\u5df2\u7ecf\u6309\u7167\u57fa\u7840\u6559\u7a0b，\u5728\u670d\u52a1\u5668\u4e0a\u62c9\u53d6\u4e86MogFace\u7684\u955c\u50cf\u6216\u4ee3\u7801\u3002\u6211\u4eec\u73b0\u5728\u8981\u505a\u7684\u4e0d\u662f\u201c\u5b89\u88c5\u201d，\u800c\u662f\u201c\u7cbe\u7b80\u201d\u3002<\/p>\n
\u9996\u5148，\u767b\u5f55\u4f60\u7684\u670d\u52a1\u5668，\u770b\u770b\u5f53\u524d\u8d44\u6e90\u4f7f\u7528\u60c5\u51b5\u3002<\/p>\n
# \u67e5\u770b\u5185\u5b58\u4f7f\u7528\u60c5\u51b5
\nfree -h<\/p>\n
# \u67e5\u770bCPU\u6838\u5fc3\u6570
\nnproc<\/p>\n
# \u67e5\u770b\u5df2\u542f\u52a8\u7684\u4e0d\u5fc5\u8981\u670d\u52a1（\u6839\u636e\u4f60\u7684\u7cfb\u7edf，\u547d\u4ee4\u53ef\u80fd\u4e0d\u540c）
\n# \u5bf9\u4e8e\u4f7f\u7528systemd\u7684\u7cfb\u7edf（\u5982Ubuntu 18.04+， CentOS 7+）
\nsudo systemctl list-units –type=service –state=running<\/p>\n
\u76ee\u6807：\u786e\u4fdd\u5728\u542f\u52a8\u4eba\u8138\u68c0\u6d4b\u670d\u52a1\u524d，\u7cfb\u7edf\u7a7a\u95f2\u5185\u5b58\u81f3\u5c11\u8fd8\u67091.2GB\u4ee5\u4e0a\u3002\u5982\u679c\u5176\u4ed6\u670d\u52a1\u5360\u7528\u4e86\u8fc7\u591a\u5185\u5b58，\u8003\u8651\u6682\u65f6\u505c\u7528\u5b83\u4eec（\u5982\u4e0d\u5fc5\u8981\u7684\u6570\u636e\u5e93\u3001docker\u670d\u52a1\u7b49）\u3002<\/p>\n
\u901a\u5e38\u7684pip install -r requirements.txt\u53ef\u80fd\u4f1a\u5b89\u88c5\u4e00\u4e9b\u6211\u4eec\u7528\u4e0d\u5230\u7684\u5927\u578b\u5e93（\u5982\u67d0\u4e9b\u7248\u672c\u7684TensorFlow\u6216\u5b8c\u6574\u7684OpenCV）\u3002\u6211\u4eec\u53ef\u4ee5\u8fdb\u884c\u9009\u62e9\u6027\u5b89\u88c5\u3002<\/p>\n
# \u8fdb\u5165\u9879\u76ee\u76ee\u5f55，\u5047\u8bbe\u662f \/root\/cv_resnet101_face-detection_cvpr22papermogface
\ncd \/root\/cv_resnet101_face-detection_cvpr22papermogface<\/p>\n
# \u521b\u5efa\u4e00\u4e2a\u7cbe\u7b80\u7248\u7684\u4f9d\u8d56\u6587\u4ef6
\ncat > requirements_light.txt << EOF
\ntorch>=1.7.0,<2.0.0 # \u6838\u5fc3\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6，\u6ce8\u610f\u7248\u672c
\ntorchvision
\nopencv-python-headless==4.5.5.64 # \u4f7f\u7528headless\u7248\u672c，\u8282\u7701\u7a7a\u95f4
\ngradio>=3.0.0,<4.0.0 # WebUI\u6846\u67b6
\nnumpy
\npillow # \u56fe\u7247\u5904\u7406
\nrequests # \u7528\u4e8eAPI\u5065\u5eb7\u68c0\u67e5
\nEOF<\/p>\n
# \u4f7f\u7528\u6e05\u534e\u955c\u50cf\u6e90\u52a0\u901f\u5b89\u88c5，\u5e76\u907f\u514d\u5b89\u88c5\u989d\u5916\u4f9d\u8d56
\npip install -i https:\/\/pypi.tuna.tsinghua.edu.cn\/simple -r requirements_light.txt –no-deps
\n# \u5982\u679c\u51fa\u73b0\u4f9d\u8d56\u7f3a\u5931\u9519\u8bef，\u518d\u5355\u72ec\u5b89\u88c5\u7f3a\u5931\u7684\u5305<\/p>\n
\u5173\u952e\u70b9：<\/p>\n
\u9ed8\u8ba4\u7684\u542f\u52a8\u547d\u4ee4\u53ef\u80fd\u4e3a\u6027\u80fd\u505a\u4e86\u59a5\u534f\u3002\u6211\u4eec\u9700\u8981\u8c03\u6574Gradio\u7684\u542f\u52a8\u53c2\u6570\u3002<\/p>\n
\u627e\u5230\u542f\u52a8WebUI\u7684\u811a\u672c\u6216\u547d\u4ee4，\u901a\u5e38\u662fpython app.py\u6216gradio app.py\u3002\u6211\u4eec\u9700\u8981\u4fee\u6539\u5176\u542f\u52a8\u65b9\u5f0f\u3002<\/p>\n
\u521b\u5efa\u4e00\u4e2a\u65b0\u7684\u542f\u52a8\u811a\u672crun_lightweight.sh：<\/p>\n
#!\/bin\/bash
\ncd \/root\/cv_resnet101_face-detection_cvpr22papermogface<\/p>\n
# \u8bbe\u7f6e\u73af\u5883\u53d8\u91cf，\u9650\u5236PyTorch\u7684\u7ebf\u7a0b\u6570，\u907f\u514d\u5360\u6ee1CPU\u5f71\u54cd\u7cfb\u7edf
\nexport OMP_NUM_THREADS=1
\nexport MKL_NUM_THREADS=1<\/p>\n
# \u542f\u52a8Gradio\u5e94\u7528，\u5173\u952e\u53c2\u6570\u5982\u4e0b：
\n# –max-file-size: \u9650\u5236\u4e0a\u4f20\u6587\u4ef6\u5927\u5c0f，\u4f8b\u598220MB，\u9632\u6b62\u5927\u56fe\u7247\u5403\u5149\u5185\u5b58
\n# –concurrency-count: \u8bbe\u7f6e\u5e76\u53d1\u5904\u7406\u6570，\u4f4e\u914d\u670d\u52a1\u5668\u5efa\u8bae\u8bbe\u4e3a1，\u540c\u4e00\u65f6\u95f4\u53ea\u5904\u7406\u4e00\u4e2a\u8bf7\u6c42
\n# –server-name: \u7ed1\u5b9a\u5230\u6240\u6709\u7f51\u7edc\u63a5\u53e3
\n# –server-port: \u6307\u5b9a\u7aef\u53e3
\n# \u5047\u8bbe\u4f60\u7684\u4e3b\u7a0b\u5e8f\u6587\u4ef6\u662f webui.py
\npython webui.py \\\\
\n –max-file-size 20 \\\\
\n –concurrency-count 1 \\\\
\n –server-name 0.0.0.0 \\\\
\n –server-port 7860<\/p>\n
\u7ed9\u811a\u672c\u6dfb\u52a0\u6267\u884c\u6743\u9650\u5e76\u8fd0\u884c：<\/p>\n
chmod +x run_lightweight.sh
\n.\/run_lightweight.sh<\/p>\n
\u53c2\u6570\u89e3\u91ca：<\/p>\n
\u670d\u52a1\u80fd\u8dd1\u8d77\u6765\u53ea\u662f\u7b2c\u4e00\u6b65，\u8981\u8dd1\u5f97\u7a33\u3001\u8dd1\u5f97\u597d，\u8fd8\u9700\u8981\u5bf9\u6a21\u578b\u52a0\u8f7d\u548c\u56fe\u7247\u5904\u7406\u6d41\u7a0b\u52a8\u624b\u672f\u3002<\/p>\n
\u9ed8\u8ba4\u60c5\u51b5\u4e0b，\u6a21\u578b\u53ef\u80fd\u4f1a\u5c1d\u8bd5\u4f7f\u7528GPU，\u6216\u8005\u5728CPU\u4e0a\u4f7f\u7528Float32（\u5355\u7cbe\u5ea6）\u8ba1\u7b97\u3002\u6211\u4eec\u53ef\u4ee5\u5f3a\u5236\u5176\u4f7f\u7528CPU\u5e76\u5c1d\u8bd5\u66f4\u4f4e\u7cbe\u5ea6\u7684Float16，\u4ee5\u8282\u7701\u5185\u5b58\u548c\u8ba1\u7b97\u91cf\u3002<\/p>\n
\u4f60\u9700\u8981\u627e\u5230\u6a21\u578b\u52a0\u8f7d\u7684\u4ee3\u7801\u90e8\u5206（\u901a\u5e38\u5728detector.py\u6216\u7c7b\u4f3c\u6587\u4ef6\u4e2d）\u3002\u4fee\u6539\u7684\u601d\u8def\u5982\u4e0b：<\/p>\n
# \u539f\u4ee3\u7801\u53ef\u80fd\u7c7b\u4f3c\u8fd9\u6837
\nimport torch
\nfrom models.mogface import MogFaceDetector<\/p>\n
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
\nmodel = MogFaceDetector(…).to(device)
\nmodel.load_state_dict(torch.load('mogface_resnet101.pth'))
\nmodel.eval()<\/p>\n
# \u4f18\u5316\u540e\u7684\u4ee3\u7801
\nimport torch
\nfrom models.mogface import MogFaceDetector<\/p>\n
# 1. \u5f3a\u5236\u4f7f\u7528CPU
\ndevice = torch.device('cpu')<\/p>\n
# 2. \u52a0\u8f7d\u6a21\u578b\u6743\u91cd\u5230CPU
\nmodel = MogFaceDetector(…)
\nstate_dict = torch.load('mogface_resnet101.pth', map_location='cpu')
\nmodel.load_state_dict(state_dict)<\/p>\n
# 3. \u5c1d\u8bd5\u8f6c\u6362\u4e3a\u534a\u7cbe\u5ea6 (Float16)，\u5982\u679c\u6a21\u578b\u652f\u6301\u7684\u8bdd
\n# \u6ce8\u610f：\u5e76\u975e\u6240\u6709\u6a21\u578b\u64cd\u4f5c\u90fd\u5b8c\u5168\u652f\u6301Float16，\u5982\u679c\u8f6c\u6362\u540e\u51fa\u9519\u6216\u7cbe\u5ea6\u635f\u5931\u4e25\u91cd，\u5219\u8df3\u8fc7\u6b64\u6b65
\ntry:
\n model = model.half() # \u8f6c\u6362\u4e3a\u534a\u7cbe\u5ea6
\n print("\u6a21\u578b\u5df2\u8f6c\u6362\u4e3a\u534a\u7cbe\u5ea6(Float16)\u6a21\u5f0f\u3002")
\nexcept Exception as e:
\n print(f"\u6a21\u578b\u4e0d\u652f\u6301\u534a\u7cbe\u5ea6\u8f6c\u6362，\u5c06\u4f7f\u7528\u5355\u7cbe\u5ea6(Float32)\u3002\u9519\u8bef: {e}")
\n pass<\/p>\n
model.to(device)
\nmodel.eval()<\/p>\n
# 4. \u5728\u63a8\u7406\u65f6，\u786e\u4fdd\u8f93\u5165\u6570\u636e\u4e5f\u5728CPU\u4e0a，\u5e76\u4e14\u7cbe\u5ea6\u5339\u914d
\ndef detect(image_tensor):
\n with torch.no_grad():
\n # \u5c06\u56fe\u7247\u6570\u636e\u4e5f\u653e\u5230CPU，\u5e76\u5339\u914d\u6a21\u578b\u7cbe\u5ea6
\n if next(model.parameters()).dtype == torch.float16:
\n image_tensor = image_tensor.half()
\n image_tensor = image_tensor.to(device)
\n predictions = model(image_tensor)
\n return predictions<\/p>\n
\u91cd\u8981\u63d0\u793a：model.half()\u8fd9\u4e00\u6b65\u9700\u8981\u6d4b\u8bd5\u3002\u534a\u7cbe\u5ea6\u80fd\u51cf\u5c11\u8fd1\u4e00\u534a\u7684\u5185\u5b58\u5360\u7528，\u4f46\u53ef\u80fd\u5bfc\u81f4\u6570\u503c\u4e0d\u7a33\u5b9a，\u5f71\u54cd\u68c0\u6d4b\u7cbe\u5ea6\u3002\u5982\u679c\u53d1\u73b0\u8f6c\u6362\u540e\u68c0\u6d4b\u6548\u679c\u53d8\u5dee\u6216\u62a5\u9519，\u8bf7\u79fb\u9664\u8fd9\u884c\u4ee3\u7801，\u4fdd\u7559Float32\u6a21\u5f0f\u3002<\/p>\n
\u56fe\u7247\u5728\u9001\u5165\u6a21\u578b\u524d，\u4f1a\u88ab\u9884\u5904\u7406（\u7f29\u653e\u3001\u5f52\u4e00\u5316\u7b49）\u3002\u5904\u7406\u8d85\u5927\u539f\u56fe\u4f1a\u6d88\u8017\u5927\u91cf\u5185\u5b58\u3002\u6211\u4eec\u53ef\u4ee5\u5728\u9884\u5904\u7406\u9636\u6bb5\u5c31\u8fdb\u884c\u9650\u5236\u3002<\/p>\n
\u5728\u4f60\u7684\u56fe\u7247\u9884\u5904\u7406\u51fd\u6570\u4e2d（\u53ef\u80fd\u5728utils.py\u6216preprocess.py），\u6dfb\u52a0\u5c3a\u5bf8\u9650\u5236：<\/p>\n
import cv2
\nfrom PIL import Image<\/p>\n
def preprocess_image_for_low_memory(image_path, max_side_length=1024):
\n """
\n \u4e3a\u4f4e\u5185\u5b58\u73af\u5883\u4f18\u5316\u7684\u56fe\u7247\u9884\u5904\u7406\u3002
\n \u5c06\u56fe\u7247\u7684\u957f\u8fb9\u7f29\u653e\u5230 max_side_length \u4ee5\u5185，\u4fdd\u6301\u5bbd\u9ad8\u6bd4\u3002
\n """
\n # \u4f7f\u7528PIL\u6216OpenCV\u8bfb\u53d6\u56fe\u7247
\n img = cv2.imread(image_path)
\n if img is None:
\n # \u5982\u679ccv2\u8bfb\u5931\u8d25，\u5c1d\u8bd5PIL
\n img = Image.open(image_path)
\n img = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)<\/p>\n
height, width = img.shape[:2]
\n max_dim = max(height, width)<\/p>\n
# \u5982\u679c\u56fe\u7247\u592a\u5927，\u8fdb\u884c\u7f29\u653e
\n if max_dim > max_side_length:
\n scale = max_side_length \/ max_dim
\n new_width = int(width * scale)
\n new_height = int(height * scale)
\n img = cv2.resize(img, (new_width, new_height), interpolation=cv2.INTER_LINEAR)
\n print(f"\u56fe\u7247\u5df2\u4ece ({width}, {height}) \u7f29\u653e\u81f3 ({new_width}, {new_height})")<\/p>\n
# \u540e\u7eed\u7684\u5f52\u4e00\u5316\u7b49\u64cd\u4f5c…
\n # … [\u4f60\u7684\u539f\u6709\u9884\u5904\u7406\u4ee3\u7801] …<\/p>\n
return processed_tensor<\/p>\n
\u540c\u65f6，\u5728WebUI\u7684\u6279\u91cf\u5904\u7406\u529f\u80fd\u4e2d，\u4e25\u683c\u9650\u5236\u4e00\u6b21\u5904\u7406\u7684\u56fe\u7247\u6570\u91cf\u3002\u4fee\u6539\u6279\u91cf\u5904\u7406\u903b\u8f91，\u4f8b\u5982\u4e00\u6b21\u6700\u591a\u5904\u74065\u5f20\u56fe\u7247，\u5e76\u5728\u524d\u7aef\u7ed9\u51fa\u63d0\u793a\u3002<\/p>\n
Gradio\u754c\u9762\u9ed8\u8ba4\u53ef\u80fd\u5305\u542b\u4e00\u4e9b\u52a8\u753b\u3001\u9ad8\u7ea7\u7ec4\u4ef6\u3002\u6211\u4eec\u53ef\u4ee5\u7b80\u5316\u754c\u9762，\u51cf\u5c11\u524d\u7aef\u8d44\u6e90\u6d88\u8017\u3002\u4fee\u6539\u4f60\u7684webui.py\u4e2d\u521b\u5efa\u754c\u9762\u7684\u90e8\u5206：<\/p>\n
import gradio as gr<\/p>\n
# \u521b\u5efa\u754c\u9762\u65f6，\u4f7f\u7528\u66f4\u7b80\u5355\u7684\u7ec4\u4ef6
\nwith gr.Blocks(title="\u8f7b\u91cf\u7ea7\u4eba\u8138\u68c0\u6d4b", theme=gr.themes.Base()) as demo: # \u4f7f\u7528\u57fa\u7840\u4e3b\u9898
\n gr.Markdown("# 🎯 \u8f7b\u91cf\u7ea7\u4eba\u8138\u68c0\u6d4b\u670d\u52a1")
\n with gr.Row():
\n with gr.Column():
\n input_image = gr.Image(label="\u4e0a\u4f20\u56fe\u7247", type="filepath") # \u4f7f\u7528\u6587\u4ef6\u8def\u5f84\u7c7b\u578b，\u6709\u65f6\u6bd4numpy\u6570\u7ec4\u66f4\u7701\u5185\u5b58
\n conf_threshold = gr.Slider(0, 1, value=0.5, label="\u7f6e\u4fe1\u5ea6\u9608\u503c")
\n btn = gr.Button("\u5f00\u59cb\u68c0\u6d4b", variant="primary")
\n with gr.Column():
\n output_image = gr.Image(label="\u68c0\u6d4b\u7ed3\u679c")
\n output_json = gr.JSON(label="\u4eba\u8138\u6570\u636e")<\/p>\n
# … \u7ed1\u5b9a\u5904\u7406\u51fd\u6570 …<\/p>\n
# \u542f\u52a8\u65f6\u4f7f\u7528\u6211\u4eec\u7684\u8f7b\u91cf\u53c2\u6570
\nif __name__ == "__main__":
\n # \u8fd9\u4e9b\u53c2\u6570\u4e5f\u53ef\u4ee5\u901a\u8fc7\u547d\u4ee4\u884c\u4f20\u5165
\n demo.launch(
\n server_name="0.0.0.0",
\n server_port=7860,
\n max_file_size="20MB", # \u9650\u5236\u4e0a\u4f20
\n concurrency_count=1, # \u9650\u5236\u5e76\u53d1
\n show_error=True,
\n quiet=True, # \u51cf\u5c11\u63a7\u5236\u53f0\u8f93\u51fa
\n share=False # \u4f4e\u914d\u670d\u52a1\u5668\u4e0d\u8981\u5f00\u542fshare\u94fe\u63a5
\n )<\/p>\n
\u670d\u52a1\u8dd1\u8d77\u6765\u540e，\u6211\u4eec\u9700\u8981\u786e\u4fdd\u5b83\u80fd\u5728\u538b\u529b\u4e0b\u4fdd\u6301\u7a33\u5b9a\u3002<\/p>\n
\u5728\u53e6\u4e00\u4e2a\u7ec8\u7aef\u7a97\u53e3，\u8fd0\u884c\u76d1\u63a7\u547d\u4ee4：<\/p>\n
# \u52a8\u6001\u67e5\u770b\u5185\u5b58\u548cCPU\u4f7f\u7528\u60c5\u51b5
\nhtop # \u5982\u679c\u672a\u5b89\u88c5，\u4f7f\u7528 `apt install htop` \u6216 `yum install htop`
\n# \u6216\u8005\u4f7f\u7528\u7b80\u5355\u7684top
\ntop<\/p>\n
# \u67e5\u770b\u7279\u5b9a\u8fdb\u7a0b\u7684\u8d44\u6e90\u4f7f\u7528（\u627e\u5230\u4f60\u7684Python\u8fdb\u7a0bPID）
\nps aux | grep python
\n# \u5047\u8bbePID\u662f 12345
\ntop -p 12345<\/p>\n
\u89c2\u5bdf\u5728\u5355\u6b21\u68c0\u6d4b\u8fc7\u7a0b\u4e2d，\u5185\u5b58\u4f7f\u7528\u91cf\u7684\u5cf0\u503c\u3002\u7406\u60f3\u60c5\u51b5\u4e0b，\u5cf0\u503c\u5e94\u63a7\u5236\u57281.5GB\u4ee5\u5185，\u7ed9\u7cfb\u7edf\u7559\u51fa\u51e0\u767eMB\u7684\u7f13\u51b2\u7a7a\u95f4\u3002<\/p>\n
\u5982\u679c\u670d\u52a1\u56e0\u4e3a\u4e0d\u53ef\u9884\u77e5\u7684\u539f\u56e0\u5d29\u6e83，\u6211\u4eec\u9700\u8981\u5b83\u80fd\u81ea\u52a8\u91cd\u542f\u3002\u53ef\u4ee5\u7528\u7b80\u5355\u7684while\u5faa\u73af\u811a\u672c，\u6216\u8005\u66f4\u597d\u7684\u5de5\u5177\u5982supervisor\u3002<\/p>\n
\u521b\u5efa\u4e00\u4e2a\u5b88\u62a4\u811a\u672cdaemon.sh：<\/p>\n
#!\/bin\/bash
\ncd \/root\/cv_resnet101_face-detection_cvpr22papermogface<\/p>\n
while true; do
\n echo "$(date): \u542f\u52a8\u4eba\u8138\u68c0\u6d4b\u670d\u52a1…"
\n # \u8fd9\u91cc\u8c03\u7528\u6211\u4eec\u4e4b\u524d\u5199\u7684\u8f7b\u91cf\u542f\u52a8\u811a\u672c
\n .\/run_lightweight.sh<\/p>\n
# \u5982\u679c\u4e0a\u9762\u7684\u811a\u672c\u9000\u51fa（\u670d\u52a1\u5d29\u6e83），\u4f1a\u6267\u884c\u5230\u8fd9\u91cc
\n echo "$(date): \u670d\u52a1\u610f\u5916\u9000\u51fa，10\u79d2\u540e\u91cd\u542f…"
\n sleep 10
\ndone<\/p>\n
\u7136\u540e\u4f7f\u7528nohup\u5728\u540e\u53f0\u8fd0\u884c\u5b83：<\/p>\n
chmod +x daemon.sh
\nnohup .\/daemon.sh > service.log 2>&1 &<\/p>\n
\u8fd9\u6837\u5373\u4f7f\u670d\u52a1\u5d29\u6e83，\u4e5f\u4f1a\u572810\u79d2\u540e\u91cd\u542f\u3002\u65e5\u5fd7\u4f1a\u8bb0\u5f55\u5728service.log\u6587\u4ef6\u4e2d\u3002<\/p>\n
\u5728\u4f4e\u914d\u670d\u52a1\u5668\u4e0a，\u6211\u4eec\u9700\u8981\u63a5\u53d7\u4e00\u4e9b\u6298\u8877：<\/p>\n
\u901a\u8fc7\u4ee5\u4e0a\u6b65\u9aa4，\u6211\u4eec\u6210\u529f\u5730\u5c06\u4e00\u4e2a\u9700\u8981\u4e00\u5b9a\u7b97\u529b\u7684MogFace\u4eba\u8138\u68c0\u6d4b\u6a21\u578b\u53ca\u5176WebUI，\u9002\u914d\u5230\u4e86\u4ec52\u68382GB\u7684\u4f4e\u914d\u670d\u52a1\u5668\u4e0a\u3002\u6211\u4eec\u6765\u56de\u987e\u4e00\u4e0b\u5173\u952e\u70b9：<\/p>\n
\u8fd9\u5957\u65b9\u6848\u7684\u610f\u4e49\u5728\u4e8e，\u5b83\u6781\u5927\u5730\u964d\u4f4e\u4e86AI\u6280\u672f\u843d\u5730\u7684\u95e8\u69db\u3002\u4f60\u53ef\u4ee5\u5c06\u8fd9\u6837\u7684\u670d\u52a1\u90e8\u7f72\u5728\u6811\u8393\u6d3e\u3001\u8001\u65e7\u7b14\u8bb0\u672c\u3001\u6216\u8005\u6700\u4fbf\u5b9c\u7684\u4e91\u670d\u52a1\u5668\u4e0a，\u7528\u4e8e\u4e2a\u4eba\u9879\u76ee\u3001\u539f\u578b\u9a8c\u8bc1\u3001\u4f4e\u5e76\u53d1\u91cf\u7684\u5185\u90e8\u5de5\u5177\u7b49\u573a\u666f\u3002<\/p>\n
\u6700\u540e\u7684\u5c0f\u5efa\u8bae：\u5728\u5b9e\u9645\u4f7f\u7528\u4e2d，\u5982\u679c\u4e3b\u8981\u901a\u8fc7API\u8c03\u7528，\u53ef\u4ee5\u8003\u8651\u4f7f\u7528\u66f4\u8f7b\u91cf\u7684Web\u6846\u67b6（\u5982FastAPI）\u66ff\u4ee3Gradio，\u5e76\u5173\u95edWebUI\u754c\u9762，\u8fd9\u6837\u53ef\u4ee5\u8fdb\u4e00\u6b65\u8282\u7701\u8d44\u6e90\u3002\u4f46\u5bf9\u4e8e\u9700\u8981\u53ef\u89c6\u5316\u64cd\u4f5c\u7684\u60c5\u51b5，\u672c\u6587\u63d0\u4f9b\u7684Gradio\u65b9\u6848\u662f\u4e00\u4e2a\u5f88\u597d\u7684\u5e73\u8861\u70b9\u3002<\/p>\n
\u5e0c\u671b\u8fd9\u4efd\u6307\u5357\u80fd\u5e2e\u52a9\u4f60，\u8ba9\u6709\u9650\u7684\u7b97\u529b\u4e5f\u80fd\u53d1\u6325\u51faAI\u7684\u4ef7\u503c\u3002<\/p>\n
\u83b7\u53d6\u66f4\u591aAI\u955c\u50cf<\/p>\n
\u60f3\u63a2\u7d22\u66f4\u591aAI\u955c\u50cf\u548c\u5e94\u7528\u573a\u666f？\u8bbf\u95ee CSDN\u661f\u56fe\u955c\u50cf\u5e7f\u573a，\u63d0\u4f9b\u4e30\u5bcc\u7684\u9884\u7f6e\u955c\u50cf，\u8986\u76d6\u5927\u6a21\u578b\u63a8\u7406\u3001\u56fe\u50cf\u751f\u6210\u3001\u89c6\u9891\u751f\u6210\u3001\u6a21\u578b\u5fae\u8c03\u7b49\u591a\u4e2a\u9886\u57df，\u652f\u6301\u4e00\u952e\u90e8\u7f72\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"
MogFace\u4eba\u8138\u68c0\u6d4b\u6a21\u578b-WebUI\u7b97\u529b\u9002\u914d\u6307\u5357：\u4f4e\u914d\u670d\u52a1\u5668（2\u68382GB）\u8f7b\u91cf\u8fd0\u884c\u6280\u5de7
\n1. \u5f15\u8a00：\u5f53\u5f3a\u5927\u6a21\u578b\u9047\u4e0a\u6709\u9650\u7b97\u529b
\n\u60f3\u8c61\u4e00\u4e0b，\u4f60\u624b\u5934\u6709\u4e00\u53f0\u95f2\u7f6e\u7684\u65e7\u670d\u52a1\u5668，\u6216\u8005\u4e00\u4e2a\u914d\u7f6e\u4e0d\u9ad8\u7684\u4e91\u4e3b\u673a，\u53ea\u67092\u4e2aCPU\u6838\u5fc3\u548c2GB\u5185\u5b58\u3002\u4f60\u60f3\u5728\u4e0a\u9762\u90e8\u7f72\u4e00\u4e2a\u5148\u8fdb\u7684\u4eba\u8138\u68c0\u6d4b\u670d\u52a1，\u7528\u6765\u5904\u7406\u56fe\u7247\u3001\u8bc6\u522b\u7528\u6237，\u751a\u81f3\u505a\u4e00\u4e9b\u7b80\u5355\u7684\u5b89\u9632\u76d1\u63a7\u3002\u4f46\u4e00\u770b\u5230\u90a3\u4e9b\u52a8\u8f84\u8981\u6c428GB\u300116GB\u5185\u5b58<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[7591,9349,5961,523],"topic":[],"class_list":["post-83503","post","type-post","status-publish","format-standard","hentry","category-server","tag-ai","tag-mogface","tag-5961","tag-523"],"yoast_head":"\n