{"id":112433,"date":"2026-10-03T15:49:01","date_gmt":"2026-10-03T07:49:01","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/112433.html"},"modified":"2026-10-03T15:49:01","modified_gmt":"2026-10-03T07:49:01","slug":"yolov13%e6%94%b9%e8%bf%9b%e7%ad%96%e7%95%a5%e3%80%90neck%e7%af%87%e3%80%91-afpn-%e6%b8%90%e8%bf%9b%e5%bc%8f%e7%89%b9%e5%be%81%e9%87%91%e5%ad%97%e5%a1%94%ef%bc%8c%e6%95%b4%e4%bd%93%e6%9b%bf%e6%8d%a2-h","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/112433.html","title":{"rendered":"YOLOv13\u6539\u8fdb\u7b56\u7565\u3010Neck\u7bc7\u3011| AFPN \u6e10\u8fdb\u5f0f\u7279\u5f81\u91d1\u5b57\u5854\uff0c\u6574\u4f53\u66ff\u6362 Head \u53c2\u6570\u53cd\u964d 82 \u4e07"},"content":{"rendered":"

\u672c\u6587\u57fa\u4e8e YOLOv13 \u5b98\u65b9\u4ed3\u5e93（iMoonLab\/yolov13，ultralytics 8.3.63 fork） \u5b9e\u6d4b\u6574\u7406，Windows\/CPU \u5168\u7a0b\u53ef\u8dd1\u3002Neck \u7bc7\u6536\u5b98\u4e4b\u4f5c\u662f\u6700\u5927\u5200\u9614\u65a7\u7684\u4e00\u7bc7：AFPN（Asymptotic Feature Pyramid Network，Yang et al., 2023，\u5b98\u65b9\u6e90\u7801 github.com\/gyyang23\/AFPN）\u4e3b\u5f20\u653e\u5f03"\u81ea\u9876\u5411\u4e0b+\u81ea\u5e95\u5411\u4e0a"\u7684\u56fa\u5b9a\u8def\u5f84，\u8ba9\u76f8\u90bb\u5c3a\u5ea6\u5148\u878d\u5408\u3001\u9010\u7ea7\u6e10\u8fdb\u3002\u672c\u6587\u7528\u5b98\u65b9 AFPN \u6574\u4f53\u66ff\u6362 v13 \u7684 HyperACE\/FullPAD Head\u2014\u2014\u5b9e\u6d4b\u53c2\u6570\u51cf\u5c11 825,746\u3001GFLOPs 7.1\u3002<\/p>\n

\u524d\u8a00<\/h3>\n

Neck \u7bc7\u6536\u5b98\u3002\u524d\u56db\u7bc7\u90fd\u662f"\u8282\u70b9\u7ea7"\u6539\u52a8（\u6362\u4e00\u4e2a\u6c47\u5408\u70b9\u3001\u4e00\u5904\u4e0a\u91c7\u6837\u3001\u4e00\u5904\u5377\u79ef）；\u8fd9\u7bc7\u662f\u7ed3\u6784\u7ea7\u6539\u52a8：\u628a v13 \u7684\u6574\u4e2a head（HyperACE \u8d85\u56fe + 7 \u4e2a FullPAD \u95e8\u63a7 + PANet \u8def\u5f84，\u5171 24 \u5c42）\u66ff\u6362\u4e3a\u4e00\u4e2a AFPN \u6e10\u8fdb\u5f0f\u878d\u5408\u6a21\u5757\u3002AFPN \u51fa\u81ea\u8bba\u6587\u300aAFPN: Asymptotic Feature Pyramid Network for Object Detection\u300b（2023），\u6838\u5fc3\u601d\u60f3\u662f\u6e10\u8fd1（asymptotic）：\u76f4\u63a5\u878d\u5408\u8bed\u4e49\u5dee\u8ddd\u5927\u7684\u76f8\u90bb\u5c3a\u5ea6\u4f1a\u6709\u4fe1\u606f\u51b2\u7a81，\u5148\u628a\u6700\u76f8\u90bb\u7684\u5c42\u7ea7\u878d\u5408\u51fa"\u4e2d\u95f4\u7ed3\u679c"，\u518d\u9010\u7ea7\u5411\u66f4\u8fdc\u5c3a\u5ea6\u6269\u6563\u3002\u66ff\u6362\u540e\u6574\u7f51\u53ea\u6709 14 \u5c42\u3002<\/p>\n

\u4e13\u680f\u76ee\u5f55：YOLOv13\u6539\u8fdb\u76ee\u5f55\u4e00\u89c8<\/p>\n

\u4e13\u680f\u5730\u5740：YOLOv13\u6539\u8fdb\u4e13\u680f\u2014\u2014\u6301\u7eed\u66f4\u65b0\u5404\u65b9\u5411\u5373\u63d2\u5373\u7528\u6539\u8fdb<\/p>\n

\u4e00\u3001\u66ff\u6362\u70b9\u5206\u6790：\u6574\u4e2a Head<\/h3>\n

backbone:
\n – [-1, 1, Conv, [64, 3, 2]] # 0-P1\/2
\n – [-1, 1, Conv, [128, 3, 2, 1, 2]] # 1-P2\/4
\n – [-1, 2, DSC3k2, [256, False, 0.25]]# 2 \u2190 AFPN \u7684\u7b2c 4 \u8def\u8f93\u5165（P2 \u5c3a\u5ea6，64 \u901a\u9053）
\n – [-1, 1, Conv, [256, 3, 2, 1, 4]] # 3-P3\/8
\n – [-1, 2, DSC3k2, [512, False, 0.25]]# 4 \u2190 \u7b2c 1 \u8def（P3，128 \u901a\u9053）
\n – [-1, 1, DSConv, [512, 3, 2]] # 5-P4\/16
\n – [-1, 4, A2C2f, [512, True, 4]] # 6 \u2190 \u7b2c 2 \u8def（P4，128 \u901a\u9053）
\n – [-1, 1, DSConv, [1024, 3, 2]] # 7-P5\/32
\n – [-1, 4, A2C2f, [1024, True, 1]] # 8 \u2190 \u7b2c 3 \u8def（P5，256 \u901a\u9053）<\/p>\n

head:
\n – [[2, 4, 6, 8], 1, AFPN, [128]] # 9 \u2605AFPN（\u6574\u4f53\u66ff\u6362，4 \u5c3a\u5ea6\u6e10\u8fdb\u878d\u5408）
\n – [9, 1, GoI, [1]] # 10 \u53d6 P3
\n – [9, 1, GoI, [2]] # 11 \u53d6 P4
\n – [9, 1, GoI, [3]] # 12 \u53d6 P5
\n – [[10, 11, 12], 1, Detect, [nc]] # 13 Detect(P3, P4, P5)<\/p>\n

\u4e24\u4e2a\u5173\u952e\u70b9：<\/p>\n

  • AFPN \u5b98\u65b9\u662f 4 \u8f93\u5165\u7ed3\u6784，\u800c v13 backbone \u6070\u597d\u6709 4 \u4e2a\u5929\u7136\u5c3a\u5ea6\u2014\u2014\u5c42 2（P2\/4，64 \u901a\u9053）\u3001\u5c42 4（P3\/8，128 \u901a\u9053）\u3001\u5c42 6（P4\/16，128 \u901a\u9053）\u3001\u5c42 8（P5\/32，256 \u901a\u9053），\u4e0e\u5b98\u65b9 [C2, C3, C4, C5] \u7684\u5c3a\u5ea6\u7ea6\u5b9a\u5b8c\u5168\u5bf9\u5e94；<\/li>\n
  • \u591a\u8f93\u51fa\u7684\u6865\u6881：AFPN \u8f93\u51fa 4 \u8def\u7279\u5f81，\u800c Detect \u9700\u8981\u6309\u5c42\u53d6\u7528\u2014\u2014\u672c\u6587\u5f15\u5165\u4e00\u4e2a 7 \u53c2\u6570\u7684\u8f85\u52a9\u6a21\u5757 GoI（Go-Index，\u53d6\u7b2c idx \u8def）\u642d\u6865\u3002yaml \u4ece 33 \u5c42\u7626\u8eab\u5230 14 \u5c42\u3002<\/li>\n

    \u4e8c\u3001AFPN \u539f\u7406<\/h3>\n
      \n
    • \u4f20\u7edf FPN\/PAN \u7684\u8def\u5f84\u662f\u5199\u6b7b\u7684：\u5148\u81ea\u9876\u5411\u4e0b\u3001\u518d\u81ea\u5e95\u5411\u4e0a，\u6240\u6709\u76f8\u90bb\u5c3a\u5ea6\u65e0\u5dee\u522b\u878d\u5408\u3002AFPN \u95ee：\u8bed\u4e49\u5dee\u8ddd\u60ac\u6b8a\u7684\u5c3a\u5ea6（P5 \u4e0e P3）\u76f4\u63a5\u878d\u5408，\u51b2\u7a81\u8c01\u6765\u89e3\u51b3？<\/li>\n
    • \u6e10\u8fd1\u878d\u5408：\u53ea\u8ba9\u76f8\u90bb\u5c3a\u5ea6\u5148\u878d\u5408（P2+P3 \u2192 P3'；P3'+P4 \u2192 P4'\u2026\u2026），\u8bed\u4e49\u5dee\u8ddd\u9010\u7ea7\u6d88\u5316\u3002\u6bcf\u4e00\u7ea7\u878d\u5408\u7528 ASFF \u8282\u70b9（\u81ea\u9002\u5e94\u7a7a\u95f4\u878d\u5408，\u9010\u50cf\u7d20 softmax \u6743\u91cd）\u5b8c\u6210；<\/li>\n
    • \u6e10\u8fdb\u7684\u4e24\u8f6e：\u7b2c\u4e00\u8f6e\u5b8c\u6210\u57fa\u7840\u878d\u5408，\u7b2c\u4e8c\u8f6e\u5728\u878d\u5408\u7ed3\u679c\u4e0a\u518d\u505a\u4e00\u6b21\u5e26\u6b8b\u5dee\u7684\u6e10\u8fdb\u878d\u5408，\u6700\u540e\u5404\u5c3a\u5ea6\u8fc7 BasicBlock \u7cbe\u70bc\u8f93\u51fa；<\/li>\n
    • \u5c3a\u5ea6\u5bf9\u9f50：\u8de8\u5c3a\u5ea6\u878d\u5408\u524d\u7528 1\u00d71 \u5377\u79ef（Downsample_x2\/x4\/x8 \u4e0e Upsample）\u5bf9\u9f50\u7a7a\u95f4\u5c3a\u5bf8\u4e0e\u901a\u9053\u3002<\/li>\n<\/ul>\n

      \u4e00\u53e5\u8bdd：\u628a"\u91d1\u5b57\u5854\u7684\u5f62\u72b6"\u4e5f\u4ea4\u7ed9\u7f51\u7edc\u6309\u5185\u5bb9\u7ec4\u7ec7，\u800c\u4e0d\u662f\u8bbe\u8ba1\u8005\u753b\u6b7b\u3002<\/p>\n

      \u26a0\ufe0f \u5b9e\u73b0\u8bf4\u660e：\u5b98\u65b9 AFPN \u4e3a mmdetection \u7684 4 \u7ea7\u5b9e\u73b0，\u672c\u6587\u9010\u7c7b\u5185\u5d4c（BasicConv\/BasicBlock\/Upsample\/Downsample\/ASFF_2I\/ASFF_3\/ASFF_4\/BlockBody），\u4ec5\u5c06\u5b98\u65b9\u7c7b\u540d ASFF_2 \u6539\u4e3a ASFF_2I（\u907f\u514d\u4e0e ultralytics parse_model \u7684\u5c5e\u6027\u540d\u51b2\u7a81，\u89c1\u4e0b\u65b9\u65b0\u5751）\u3002\u5185\u90e8\u8282\u70b9\u4e0e\u5b98\u65b9\u9010\u4f4d\u4e00\u81f4\u3002<\/p>\n

      \u4e09\u3001\u5b9e\u73b0\u4ee3\u7801<\/h3>\n

      \u6a21\u5757\u8f83\u957f（BasicConv \/ BasicBlock \/ Upsample \/ Downsample_x2\/x4\/x8 \/ ASFF_2I \/ ASFF_3 \/ ASFF_4 \/ BlockBody \/ AFPN \/ GoI \u5171\u7ea6 230 \u884c），\u5b8c\u6574\u4ee3\u7801\u53ef\u5728\u4e13\u680f\u914d\u5957\u4ee3\u7801\u5305\u4e2d\u83b7\u53d6，\u6838\u5fc3\u9aa8\u67b6\u5982\u4e0b：<\/p>\n

      import torch
      \nimport torch.nn as nn
      \nimport torch.nn.functional as F<\/p>\n

      class BasicConv(nn.Module):
      \n """\u5b98\u65b9 AFPN \u7684 BasicConv：Conv + BN + SiLU（k>1 \u65f6\u81ea\u52a8\u8865 padding）\u3002"""<\/p>\n

      def __init__(self, in_planes, out_planes, kernel_size, stride=1, pad=None,
      \n relu=True, bn=True, bias=False):
      \n super().__init__()
      \n if not pad:
      \n pad = (kernel_size – 1) \/\/ 2 if kernel_size else 0
      \n self.conv = nn.Conv2d(in_planes, out_planes, kernel_size, stride=stride,
      \n padding=pad, groups=groups if False else 1, bias=bias)
      \n self.bn = nn.BatchNorm2d(out_planes, eps=1e-5, momentum=0.01, affine=True) if bn else None
      \n self.relu = nn.ReLU() if relu else None<\/p>\n

      def forward(self, x):
      \n x = self.conv(x)
      \n if self.bn is not None:
      \n x = self.bn(x)
      \n if self.relu is not None:
      \n x = self.relu(x)
      \n return x<\/p>\n

      class ASFF_2(nn.Module):
      \n """\u4e24\u8f93\u5165 ASFF \u878d\u5408\u8282\u70b9：softmax \u9010\u50cf\u7d20\u52a0\u6743\u3002"""<\/p>\n

      def __init__(self, inter_dim):
      \n super().__init__()
      \n compress_c = 8
      \n self.weight_level_1 = BasicConv(inter_dim, compress_c, 1, 1)
      \n self.weight_level_2 = BasicConv(inter_dim, compress_c, 1, 1)
      \n self.weight_levels = nn.Conv2d(compress_c * 2, 2, 1, 1, 0)
      \n self.conv = BasicConv(inter_dim, inter_dim, 3, 1)<\/p>\n

      def forward(self, input1, input2):
      \n w1 = self.weight_level_1(input1)
      \n w2 = self.weight_level_2(input2)
      \n wei = self.weight_levels(torch.cat((w1, w2), 1))
      \n wei = F.softmax(wei, dim=1)
      \n return input1 * wei[:, 0:1] + input2 * wei[:, 1:2]<\/p>\n

      # Downsample_x2\/x4\/x8\u3001Upsample\u3001ASFF_3（\u4e09\u8f93\u5165）\u3001ASFF_4（\u56db\u8f93\u5165）\u3001
      \n# BlockBody（\u4e24\u8f6e\u6e10\u8fdb\u878d\u5408）\u4e0e\u5b98\u65b9\u5b9e\u73b0\u4e00\u81f4，\u7bc7\u5e45\u6240\u9650\u4ece\u7565；
      \n# \u5b8c\u6574\u4ee3\u7801\u4e0e\u4e13\u680f\u914d\u5957\u5305\u4e00\u81f4\u3002<\/p>\n

      class AFPN(nn.Module):
      \n """AFPN \u6e10\u8fdb\u5f0f\u7279\u5f81\u91d1\u5b57\u5854：\u8f93\u5165 4 \u4e2a\u5c3a\u5ea6，\u8f93\u51fa 4 \u8def\u878d\u5408\u7279\u5f81\u3002"""<\/p>\n

      def __init__(self, in_channels, out_channels=128):
      \n super().__init__()
      \n self.conv0 = BasicConv(in_channels[0], in_channels[0] \/\/ 8, 1)
      \n self.conv1 = BasicConv(in_channels[1], in_channels[1] \/\/ 8, 1)
      \n self.conv2 = BasicConv(in_channels[2], in_channels[2] \/\/ 8, 1)
      \n self.conv3 = BasicConv(in_channels[3], in_channels[3] \/\/ 8, 1)
      \n self.body = BlockBody([in_channels[0] \/\/ 8, in_channels[1] \/\/ 8,
      \n in_channels[2] \/\/ 8, in_channels[3] \/\/ 8])
      \n self.conv00 = BasicConv(in_channels[0] \/\/ 8, out_channels, 1)
      \n self.conv11 = BasicConv(in_channels[1] \/\/ 8, out_channels, 1)
      \n self.conv22 = BasicConv(in_channels[2] \/\/ 8, out_channels, 1)
      \n self.conv33 = BasicConv(in_channels[3] \/\/ 8, out_channels, 1)<\/p>\n

      def forward(self, x):
      \n x0, x1, x2, x3 = x
      \n x0 = self.conv0(x0)
      \n x1 = self.conv1(x1)
      \n x2 = self.conv2(x2)
      \n x3 = self.conv3(x3)
      \n out0, out1, out2, out3 = self.body([x0, x1, x2, x3])
      \n return self.conv00(out0), self.conv11(out1), self.conv22(out2), self.conv33(out3)<\/p>\n

      class GoI(nn.Module):
      \n """GoIndex：\u4ece\u591a\u8f93\u51fa\u6a21\u5757\u53d6\u7b2c idx \u8def\u7279\u5f81\u3002"""<\/p>\n

      def __init__(self, c1, idx=0):
      \n super().__init__()
      \n self.idx = idx<\/p>\n

      def forward(self, x):
      \n t = x[0] if isinstance(x, (list, tuple)) and len(x) == 1 else x
      \n return t[self.idx]<\/p>\n

      \u26a0\ufe0f v13 \u65b0\u5751（\u5c5e\u6027\u540d\u51b2\u7a81）：ultralytics \u7684 parse_model \u4f1a\u7ed9\u6bcf\u4e2a\u5c42\u5b9e\u4f8b\u5199\u5165 m_.f（from \u7d22\u5f15）\u7b49\u5c5e\u6027，\u5b98\u65b9\u7c7b\u91cc\u82e5\u6709\u540c\u540d\u5b50\u6a21\u5757（\u5982 self.f）\u4f1a\u76f4\u63a5 TypeError\u3002AFPN \u7684\u878d\u5408\u8282\u70b9 ASFF_2\/ASFF_4 \u65e0\u6b64\u95ee\u9898，\u4f46\u4e3a\u533a\u5206"\u5bf9\u5916\u6ce8\u518c\u7684\u5217\u8868\u8f93\u5165\u7248"\u4e0e"BlockBody \u5185\u90e8\u7684\u53cc\u53c2\u6570\u7248"，\u672c\u6587\u5c06\u5185\u90e8\u8282\u70b9\u547d\u540d\u4e3a ASFF_2I\u3002\u6ce8\u518c\u5230 ultralytics \u7684\u5b98\u65b9\u6a21\u5757，\u6210\u5458\u547d\u540d\u52a1\u5fc5\u907f\u5f00 f\/i\/type\/np\/params\u3002<\/p>\n

      \u56db\u3001\u6dfb\u52a0\u6b65\u9aa4（v13 \u7248）<\/h3>\n

      1. \u4fee\u6539 ultralytics\/nn\/modules\/conv.py<\/h4>\n

      BasicConv（\u82e5\u5df2\u6709\u5219\u8df3\u8fc7）\u3001BasicBlock\u3001Upsample\u3001Downsample_x2\/x4\/x8\u3001ASFF_2I\u3001ASFF_3\u3001ASFF_4\u3001BlockBody\u3001AFPN\u3001GoI \u4f9d\u6b21\u7c98\u8d34\u5230 conv.py \u672b\u5c3e\u3002<\/p>\n

      2. \u4fee\u6539 ultralytics\/nn\/modules\/__init__.py<\/h4>\n

      from .conv import (…) \u91cc\u8ffd\u52a0 AFPN, \u4e0e GoI,；__all__ \u91cc\u7ed9\u6700\u540e\u4e00\u9879 "DSConv" \u8865\u9017\u53f7\u540e\u8ffd\u52a0 "AFPN"\u3001"GoI"\u3002<\/p>\n

      3. \u4fee\u6539 ultralytics\/nn\/tasks.py<\/h4>\n

      \u9876\u90e8\u5bfc\u5165\u5757\u52a0 AFPN\u3001GoI；parse_model() \u91cc\u65b0\u589e\u4e24\u4e2a\u5206\u652f（AFPN \u591a\u8f93\u5165\u53d6\u901a\u9053\u5217\u8868；GoI \u53d6\u8def）：<\/p>\n

      elif m is AFPN: # AFPN：\u591a\u8f93\u5165\u6574\u4f53 Neck，\u8f93\u51fa\u901a\u9053\u53d6\u58f0\u660e\u7684\u7edf\u4e00\u503c
      \n args.insert(0, [ch[x] for x in f])
      \n c2 = args[-1]
      \n elif m is GoI: # GoI：\u4ece\u591a\u8f93\u51fa\u6a21\u5757\u53d6\u7b2c idx \u8def
      \n c2 = ch[f]
      \n args = [c2, *args]<\/p>\n

      \u63d2\u5165\u4f4d\u7f6e：elif m is FullPAD_Tunnel: \u4e4b\u540e\u3001else: \u4e4b\u524d\u3002<\/p>\n

      \u4e94\u3001yaml \u6a21\u578b\u6587\u4ef6<\/h3>\n

      \u6574\u4f53\u66ff\u6362 Head，\u5168\u7f51\u4ec5 14 \u5c42：<\/p>\n

      # Ultralytics YOLOv13n + AFPN \u6e10\u8fdb\u5f0f\u7279\u5f81\u91d1\u5b57\u5854（\u6574\u4f53\u66ff\u6362 Head）🚀
      \nnc: 80
      \nscales:
      \n n: [0.50, 0.25, 1024]
      \n s: [0.50, 0.50, 1024]
      \n l: [1.00, 1.00, 512]
      \n x: [1.00, 1.50, 512]<\/p>\n

      backbone:
      \n # [from, repeats, module, args]
      \n – [-1, 1, Conv, [64, 3, 2]] # 0-P1\/2
      \n – [-1, 1, Conv, [128, 3, 2, 1, 2]] # 1-P2\/4
      \n – [-1, 2, DSC3k2, [256, False, 0.25]]# 2（P2 \u5c3a\u5ea6，AFPN \u7b2c 4 \u8def\u8f93\u5165）
      \n – [-1, 1, Conv, [256, 3, 2, 1, 4]] # 3-P3\/8
      \n – [-1, 2, DSC3k2, [512, False, 0.25]]# 4（P3，AFPN \u7b2c 1 \u8def\u8f93\u5165）
      \n – [-1, 1, DSConv, [512, 3, 2]] # 5-P4\/16
      \n – [-1, 4, A2C2f, [512, True, 4]] # 6（P4，\u7b2c 2 \u8def\u8f93\u5165）
      \n – [-1, 1, DSConv, [1024, 3, 2]] # 7-P5\/32
      \n – [-1, 4, A2C2f, [1024, True, 1]] # 8（P5，\u7b2c 3 \u8def\u8f93\u5165）<\/p>\n

      head:
      \n – [[2, 4, 6, 8], 1, AFPN, [128]] # 9 \u2605AFPN（\u6574\u4f53\u66ff\u6362 Head）
      \n – [9, 1, GoI, [1]] # 10 \u53d6 P3
      \n – [9, 1, GoI, [2]] # 11 \u53d6 P4
      \n – [9, 1, GoI, [3]] # 12 \u53d6 P5
      \n – [[10, 11, 12], 1, Detect, [nc]] # 13 Detect(P3, P4, P5)<\/p>\n

      \u516d\u3001\u6210\u529f\u8fd0\u884c\u7ed3\u679c<\/h3>\n

      yolo detect train model=yolov13n-AFPN.yaml data=mydata.yaml epochs=100 imgsz=640 batch=16<\/p>\n

      Python \u65b9\u5f0f：<\/p>\n

      from ultralytics import YOLO<\/p>\n

      model = YOLO("yolov13n-AFPN.yaml").load("yolov13n.pt")
      \nmodel.train(data="mydata.yaml", epochs=100, imgsz=640, batch=16)<\/p>\n

      \u672c\u6587\u5b9e\u6d4b（YOLOv13 \u5b98\u65b9\u4ed3\u5e93，Windows CPU，v13n \u7ed3\u6784，\u6574\u4f53\u66ff\u6362 Head）：<\/p>\n

      \u57fa\u7ebf yolov13n: 2,494,151 parameters, 6.5 GFLOPs
      \n+AFPN: 1,668,405 parameters, 7.1 GFLOPs Head\u2192AFPN（\u5168\u7f5114\u5c42）
      \n3 epochs completed. Results saved to runs\\\\detect\\\\v13_afpn_smoke<\/p>\n

      \u2705 \u65ad\u8a00\u5168\u90e8\u901a\u8fc7：\u2460 \u5185\u5d4c\u7684\u878d\u5408\u8282\u70b9\u4e0e\u5b98\u65b9 AFPN \u5b9e\u73b0\u540c\u6743\u91cd\u8f93\u51fa\u9010\u4f4d\u4e00\u81f4（ASFF_2 max diff = 0）；\u2461 \u7b2c 9 \u5c42\u4e3a AFPN，\u603b\u53c2\u6570\u91cf\u6bd4\u57fa\u7ebf\u51cf\u5c11 825,746（HyperACE\/FullPAD \u6574\u5957\u7ed3\u6784\u9000\u573a）；\u2462 14 \u5c42 yaml \u7684 4 \u8def\u8f93\u5165\u4e0e GoI \u53d6\u8def\u5168\u90e8\u6b63\u786e，\u6574\u7f51 640 \u524d\u5411\u6b63\u5e38，3 \u8f6e\u5192\u70df\u8bad\u7ec3\u6b63\u5e38\u6536\u655b\u3002<\/p>\n\n\u6a21\u578bParamsGFLOPs\u8bf4\u660e<\/tr>\n\n\n\n
      YOLOv13n \u57fa\u7ebf<\/td>\n2,494,151<\/td>\n6.5<\/td>\nHyperACE + 7\u00d7FullPAD + PANet<\/td>\n<\/tr>\n
      +AFPN<\/td>\n1,668,405<\/td>\n7.1<\/td>\n\u6e10\u8fdb\u5f0f\u878d\u5408，\u5168\u7f51 14 \u5c42<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

      \u63d0\u793a：AFPN \u662f\u7ed3\u6784\u7ea7\u66ff\u6362，\u4e0e\u8282\u70b9\u7ea7\u6539\u8fdb\u4e0d\u662f\u540c\u4e00\u91cf\u7ea7\u7684\u5b9e\u9a8c\u2014\u2014\u5b83\u6539\u53d8\u7684\u662f"\u878d\u5408\u7684\u8def\u5f84\u7ec4\u7ec7"\u672c\u8eab\u3002GFLOPs \u7565\u5347（P2 \u9ad8\u5206\u8fa8\u7387\u8def\u53c2\u4e0e\u878d\u5408）\u4e0e\u53c2\u6570\u5927\u964d\u5e76\u5b58，\u5c5e\u4e8e"\u7528\u7b97\u529b\u6362\u7ed3\u6784\u7b80\u5316"\u7684\u5178\u578b\u3002\u82e5\u4f60\u7684\u4efb\u52a1\u5bf9\u5e73\u79fb\u7a33\u5b9a\u6027\u548c\u591a\u5c3a\u5ea6\u4e00\u81f4\u6027\u654f\u611f，\u8fd9\u5957\u6e10\u8fdb\u5f0f\u7ed3\u6784\u503c\u5f97\u5b8c\u6574\u6d88\u878d\u3002<\/p>\n

      \u4e03\u3001\u603b\u7ed3<\/h3>\n

      AFPN \u7684\u56db\u70b9\u8bb0\u5fc6\u951a：\u6e10\u8fd1\u878d\u5408（\u53ea\u878d\u5408\u76f8\u90bb\u5c3a\u5ea6，\u8bed\u4e49\u51b2\u7a81\u9010\u7ea7\u6d88\u5316）\u3001ASFF \u8282\u70b9\u662f\u878d\u5408\u7684\u539f\u5b50（\u9010\u50cf\u7d20 softmax \u6743\u91cd）\u30014 \u5c3a\u5ea6\u8f93\u5165\u6070\u597d\u5bf9\u4e0a v13 \u7684\u5c42 2\/4\/6\/8（P2 \u9ad8\u5206\u8fa8\u7387\u8def\u53d8\u5e9f\u4e3a\u5b9d）\u3001GoI \u6865\u6881（\u591a\u8f93\u51fa\u6a21\u5757\u4e0e ultralytics \u5355\u5f20\u91cf\u4e66\u5199\u7684\u901a\u7528\u89e3\u6cd5）\u3002\u5b83\u4e5f\u662f Neck \u7bc7\u7684\u65b9\u6cd5\u8bba\u6536\u5b98：\u8282\u70b9\u7ea7（BiFPN\/GSConv\/CARAFE\/ASFF_2）\u89e3\u51b3"\u67d0\u4e2a\u70b9\u600e\u4e48\u505a"，\u7ed3\u6784\u7ea7（AFPN）\u89e3\u51b3"\u6574\u6761\u8def\u600e\u4e48\u7ec4\u7ec7"\u2014\u2014\u4e24\u5c42\u601d\u8def\u914d\u5408\u4f7f\u7528，\u624d\u80fd\u628a Neck \u6539\u900f\u3002<\/p>\n

      Neck \u7bc7\u4e94\u7bc7\u5230\u6b64\u5b8c\u6574\u3002\u4e0b\u4e00\u7bc7\u8fdb\u5165\u3010\u635f\u5931\u51fd\u6570\u7bc7\u3011\u2014\u2014WIoU v3 \u52a8\u6001\u975e\u5355\u8c03\u805a\u7126\u635f\u5931，\u4e13\u6cbb\u4f4e\u8d28\u91cf\u6807\u6ce8\u3002<\/p>\n

      \u89c9\u5f97\u6709\u5e2e\u52a9\u7684\u8bdd，\u70b9\u8d5e\u6536\u85cf\u5173\u6ce8\u4e09\u8fde\u652f\u6301\u4e00\u4e0b，\u8bc4\u8bba\u533a\u6b22\u8fce\u4ea4\u6d41\u4f60\u7684\u5b9e\u9a8c\u7ed3\u679c~<\/p>\n

      \u4e13\u680f\u76ee\u5f55：YOLOv13\u6539\u8fdb\u76ee\u5f55\u4e00\u89c8 \u4e13\u680f\u5730\u5740：YOLOv13\u6539\u8fdb\u4e13\u680f\u2014\u2014\u6301\u7eed\u66f4\u65b0\u5404\u65b9\u5411\u5373\u63d2\u5373\u7528\u6539\u8fdb<\/p>\n","protected":false},"excerpt":{"rendered":"

      \u672c\u6587\u57fa\u4e8e YOLOv13 \u5b98\u65b9\u4ed3\u5e93（iMoonLab\/yolov13，ultralytics 8.3.63 fork） \u5b9e\u6d4b\u6574\u7406，Windows\/CPU \u5168\u7a0b\u53ef\u8dd1\u3002Neck \u7bc7\u6536\u5b98\u4e4b\u4f5c\u662f\u6700\u5927\u5200\u9614\u65a7\u7684\u4e00\u7bc7：AFPN（Asymptotic Feature Pyramid Network，Yang et al., 2023，\u5b98\u65b9\u6e90\u7801 github.com\/gyyang23<\/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":[156],"topic":[],"class_list":["post-112433","post","type-post","status-publish","format-standard","hentry","category-server","tag-yolo"],"yoast_head":"\nYOLOv13\u6539\u8fdb\u7b56\u7565\u3010Neck\u7bc7\u3011| AFPN \u6e10\u8fdb\u5f0f\u7279\u5f81\u91d1\u5b57\u5854\uff0c\u6574\u4f53\u66ff\u6362 Head \u53c2\u6570\u53cd\u964d 82 \u4e07 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/112433.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"YOLOv13\u6539\u8fdb\u7b56\u7565\u3010Neck\u7bc7\u3011| AFPN \u6e10\u8fdb\u5f0f\u7279\u5f81\u91d1\u5b57\u5854\uff0c\u6574\u4f53\u66ff\u6362 Head \u53c2\u6570\u53cd\u964d 82 \u4e07 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u672c\u6587\u57fa\u4e8e YOLOv13 \u5b98\u65b9\u4ed3\u5e93（iMoonLab\/yolov13，ultralytics 8.3.63 fork） \u5b9e\u6d4b\u6574\u7406，Windows\/CPU \u5168\u7a0b\u53ef\u8dd1\u3002Neck \u7bc7\u6536\u5b98\u4e4b\u4f5c\u662f\u6700\u5927\u5200\u9614\u65a7\u7684\u4e00\u7bc7：AFPN（Asymptotic Feature Pyramid Network，Yang et al., 2023，\u5b98\u65b9\u6e90\u7801 github.com\/gyyang23\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/112433.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-10-03T07:49:01+00:00\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"6 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/112433.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/112433.html\",\"name\":\"YOLOv13\u6539\u8fdb\u7b56\u7565\u3010Neck\u7bc7\u3011| AFPN \u6e10\u8fdb\u5f0f\u7279\u5f81\u91d1\u5b57\u5854\uff0c\u6574\u4f53\u66ff\u6362 Head \u53c2\u6570\u53cd\u964d 82 \u4e07 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-10-03T07:49:01+00:00\",\"dateModified\":\"2026-10-03T07:49:01+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/112433.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/112433.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/112433.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"YOLOv13\u6539\u8fdb\u7b56\u7565\u3010Neck\u7bc7\u3011| AFPN \u6e10\u8fdb\u5f0f\u7279\u5f81\u91d1\u5b57\u5854\uff0c\u6574\u4f53\u66ff\u6362 Head \u53c2\u6570\u53cd\u964d 82 \u4e07\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"YOLOv13\u6539\u8fdb\u7b56\u7565\u3010Neck\u7bc7\u3011| AFPN \u6e10\u8fdb\u5f0f\u7279\u5f81\u91d1\u5b57\u5854\uff0c\u6574\u4f53\u66ff\u6362 Head \u53c2\u6570\u53cd\u964d 82 \u4e07 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/112433.html","og_locale":"zh_CN","og_type":"article","og_title":"YOLOv13\u6539\u8fdb\u7b56\u7565\u3010Neck\u7bc7\u3011| AFPN \u6e10\u8fdb\u5f0f\u7279\u5f81\u91d1\u5b57\u5854\uff0c\u6574\u4f53\u66ff\u6362 Head \u53c2\u6570\u53cd\u964d 82 \u4e07 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u672c\u6587\u57fa\u4e8e YOLOv13 \u5b98\u65b9\u4ed3\u5e93（iMoonLab\/yolov13，ultralytics 8.3.63 fork） \u5b9e\u6d4b\u6574\u7406，Windows\/CPU \u5168\u7a0b\u53ef\u8dd1\u3002Neck \u7bc7\u6536\u5b98\u4e4b\u4f5c\u662f\u6700\u5927\u5200\u9614\u65a7\u7684\u4e00\u7bc7：AFPN（Asymptotic Feature Pyramid Network，Yang et al., 2023，\u5b98\u65b9\u6e90\u7801 github.com\/gyyang23","og_url":"https:\/\/www.wsisp.com\/helps\/112433.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-10-03T07:49:01+00:00","author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"6 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/112433.html","url":"https:\/\/www.wsisp.com\/helps\/112433.html","name":"YOLOv13\u6539\u8fdb\u7b56\u7565\u3010Neck\u7bc7\u3011| AFPN \u6e10\u8fdb\u5f0f\u7279\u5f81\u91d1\u5b57\u5854\uff0c\u6574\u4f53\u66ff\u6362 Head \u53c2\u6570\u53cd\u964d 82 \u4e07 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-10-03T07:49:01+00:00","dateModified":"2026-10-03T07:49:01+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/112433.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/112433.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/112433.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"YOLOv13\u6539\u8fdb\u7b56\u7565\u3010Neck\u7bc7\u3011| AFPN \u6e10\u8fdb\u5f0f\u7279\u5f81\u91d1\u5b57\u5854\uff0c\u6574\u4f53\u66ff\u6362 Head \u53c2\u6570\u53cd\u964d 82 \u4e07"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/112433","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=112433"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/112433\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=112433"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=112433"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=112433"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=112433"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}