{"id":110602,"date":"2026-09-29T03:48:49","date_gmt":"2026-09-28T19:48:49","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/110602.html"},"modified":"2026-09-29T03:48:49","modified_gmt":"2026-09-28T19:48:49","slug":"%e5%b7%a5%e4%b8%9a%e7%ba%a7%e9%ab%98%e7%b2%be%e5%ba%a6%e5%a4%9a%e6%a8%a1%e6%80%81%e5%9b%be%e6%96%87%e6%a3%80%e7%b4%a2%e5%9c%a8%e4%bd%8e%e5%bb%b6%e8%bf%9f%e8%be%b9%e7%bc%98%e8%ae%be%e5%a4%87%e4%b8%8a","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/110602.html","title":{"rendered":"\u5de5\u4e1a\u7ea7\u9ad8\u7cbe\u5ea6\u591a\u6a21\u6001\u56fe\u6587\u68c0\u7d22\u5728\u4f4e\u5ef6\u8fdf\u8fb9\u7f18\u8bbe\u5907\u4e0a\u7684\u91cf\u5316\u4e0e\u975e\u5bf9\u79f0\u8de8\u6a21\u6001\u84b8\u998f\u5b9e\u6218"},"content":{"rendered":"<h2>\u5de5\u4e1a\u7ea7\u9ad8\u7cbe\u5ea6\u591a\u6a21\u6001\u56fe\u6587\u68c0\u7d22\u5728\u4f4e\u5ef6\u8fdf\u8fb9\u7f18\u8bbe\u5907\u4e0a\u7684\u91cf\u5316\u4e0e\u975e\u5bf9\u79f0\u8de8\u6a21\u6001\u84b8\u998f\u5b9e\u6218<\/h2>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/09\/20260928194847-6abac49fb3e9a.webp\" alt=\"\u5c01\u9762\u4fe1\u606f\u56fe\" \/><\/p>\n<p>\u5728\u79fb\u52a8\u7aef\u76f8\u518c\u4ee5\u56fe\u641c\u56fe\u3001\u8f66\u8f7d\u591a\u6a21\u6001\u4eba\u673a\u4ea4\u4e92&#xff08;\u5982\u8bed\u97f3\u95ee\u7b54\u68c0\u7d22\u4e2d\u63a7\u5927\u5c4f\u89c6\u89c9\u76ee\u6807&#xff09;\u3001\u4ee5\u53ca\u624b\u6301\u5de1\u68c0\u7ec8\u7aef\u56fe\u6587\u67e5\u91cd\u7b49\u8fb9\u7f18\u7aef\u90e8\u7f72&#xff08;Edge Deployment&#xff09;\u4e2d&#xff0c;\u57fa\u4e8e CLIP \u67b6\u6784\u7684\u53cc\u5854\u56fe\u6587\u8de8\u6a21\u6001\u68c0\u7d22&#xff08;Dual-Tower Cross-modal Retrieval&#xff09; \u662f\u6838\u5fc3\u5e95\u5ea7\u3002<\/p>\n<p>\u7136\u800c&#xff0c;\u5728\u9762\u5bf9**\u7b97\u529b\u6781\u5176\u532e\u4e4f\u3001\u5185\u5b58\u5e26\u5bbd\u6781\u5ea6\u72ed\u7a84\u7684\u8fb9\u7f18\u82af\u7247&#xff08;\u5982\u9ad8\u901a\u9a81\u9f99 8 Gen 3\u3001Apple A18\u3001NVIDIA Jetson Orin Nano&#xff09;**\u65f6&#xff0c;\u7b97\u6cd5\u67b6\u6784\u5e08\u9762\u4e34\u7740\u6781\u5176\u6b8b\u9177\u7684\u7269\u7406\u4e0d\u5bf9\u79f0\u74f6\u9888&#xff1a;<\/p>\n<ul>\n<li>\u7269\u7406\u4e8b\u5b9e&#xff08;\u975e\u5bf9\u79f0\u8ba1\u7b97\u5f00\u9500&#xff09;&#xff1a;\u5728\u771f\u5b9e\u68c0\u7d22\u4e1a\u52a1\u6d41\u4e2d&#xff0c;\u6587\u672c\u67e5\u8be2\u7aef&#xff08;Text Query&#xff09;\u8ba1\u7b97\u9891\u6b21\u6781\u9ad8\u3001\u8981\u6c42\u5728 5 \u6beb\u79d2\u5185\u5b8c\u6210\u7f16\u7801&#xff1b;\u800c\u56fe\u50cf\u7aef\u5e93&#xff08;Image Gallery&#xff09;\u901a\u5e38\u53ef\u4ee5\u5728\u4e91\u7aef\/\u540e\u53f0\u79bb\u7ebf\u7f16\u7801\u597d&#xff1b;<\/li>\n<li>\u82e5\u76f4\u63a5\u91c7\u7528\u5bf9\u79f0\u7684\u8f7b\u91cf\u5316\u526a\u679d\u6216\u7c97\u66b4 INT8 \u91cf\u5316&#xff0c;\u6587\u672c\u5854\u4e0e\u56fe\u50cf\u5854\u7684\u9ad8\u7ef4\u8de8\u6a21\u6001\u8bed\u4e49\u5bf9\u9f50\u7a7a\u95f4&#xff08;Cross-modal Alignment Manifold&#xff09;\u4f1a\u53d1\u751f\u5267\u70c8\u626d\u66f2\u4e0e\u574d\u584c&#xff1b;<\/li>\n<li>\u5bfc\u81f4\u56fe\u6587\u68c0\u7d22\u7684 Top-1 Recall \u4ece\u4e91\u7aef\u5927\u6a21\u578b\u7684 82% \u77ac\u95f4\u65ad\u5d16\u5f0f\u66b4\u8dcc\u81f3 38%&#xff01;<\/li>\n<\/ul>\n<p>\u901a\u8fc7\u6784\u5efa\u975e\u5bf9\u79f0\u8de8\u6a21\u6001\u77e5\u8bc6\u84b8\u998f&#xff08;Asymmetric Cross-modal Knowledge Distillation&#xff09;\u4e0e\u975e\u5bf9\u79f0\u611f\u77e5\u91cf\u5316&#xff08;Asymmetric Quantization-Aware Distillation&#xff09;\u6d41\u6c34\u7ebf&#xff1a;<\/p>\n<ul>\n<li>\u8ba9\u8f7b\u91cf\u7ea7\u6587\u672c\u5b66\u751f\u6a21\u578b&#xff08;\u8f7b\u91cf 4 \u5c42 Transformer&#xff09;\u53bb\u5f3a\u884c\u5bf9\u9f50\u7b28\u91cd\u5e9e\u5927\u7684\u4e91\u7aef\u6559\u5e08\u89c6\u89c9\u6a21\u578b&#xff08;ViT-Large \/ ViT-Huge&#xff09;&#xff1b;<\/li>\n<li>\u5728\u4fdd\u7559\u56fe\u50cf\u9ad8\u7cbe\u5ea6\u8868\u5f81\u7684\u540c\u65f6&#xff0c;\u5c06\u7aef\u4fa7\u6587\u672c\u7f16\u7801\u5ef6\u8fdf\u538b\u964d\u81f3 $2\\\\text{ms}$&#xff01;<\/li>\n<\/ul>\n<p>\u672c\u6587\u7cfb\u7edf\u5256\u6790\u975e\u5bf9\u79f0\u56fe\u6587\u68c0\u7d22\u84b8\u998f\u7684\u5fae\u89c2\u4ee3\u6570\u63a8\u5bfc\u4e0e\u5de5\u4e1a\u7ea7\u843d\u5730\u5b9e\u6218\u3002<\/p>\n<p>flowchart TD<br \/>\n    subgraph \u4e91\u7aef\u5927\u6a21\u578b\u6559\u5e08\u7f51\u7edc (Teacher &#8211; \u5e9e\u5927\u9ad8\u4fdd\u771f)<br \/>\n        A1[\u6587\u672c\u67e5\u8be2 Text Query] &#8211;&gt; B1[\u5e9e\u5927\u6559\u5e08\u6587\u672c\u7f16\u7801\u5668 Text-Teacher (ViT-L\/14)]<br \/>\n        A2[\u56fe\u50cf\u5e93 Images] &#8211;&gt; B2[\u5e9e\u5927\u6559\u5e08\u89c6\u89c9\u7f16\u7801\u5668 Vision-Teacher (ViT-L\/14)]<br \/>\n        B1 &amp; B2 &#8211;&gt; C1[\u8ba1\u7b97\u9ec4\u91d1\u9ad8\u7ef4\u8de8\u6a21\u6001\u76f8\u4f3c\u5ea6\u77e9\u9635 S_teacher in R^{B x B}]<br \/>\n    end<\/p>\n<p>    subgraph \u7aef\u4fa7\u6781\u901f\u5b66\u751f\u7f51\u7edc (Student &#8211; \u6781\u81f4\u8f7b\u91cf\u975e\u5bf9\u79f0)<br \/>\n        A1 &#8211;&gt; D1[\u8f7b\u91cf\u5b66\u751f\u6587\u672c\u7f16\u7801\u5668 Text-Student (\u4ec5 4 \u5c42 INT8)]<br \/>\n        D1 &#8211;&gt; E1[\u4e0e\u5e9e\u5927\u89c6\u89c9\u7279\u5f81\u8ba1\u7b97\u5b66\u751f\u76f8\u4f3c\u5ea6 S_student]<br \/>\n    end<\/p>\n<p>    C1 &amp; E1 &#8211;&gt; F[\u975e\u5bf9\u79f0 KL \u6563\u5ea6\u4e0e\u8de8\u6a21\u6001\u5bf9\u6bd4\u84b8\u998f\u635f\u5931 (Asymmetric Distillation Loss)]<br \/>\n    F &#8211;&gt; G[\u7aef\u4fa7\u6587\u672c\u63a8\u7406\u8017\u65f6\u4ec5 1.8ms, Top-1 \u68c0\u7d22\u53ec\u56de\u7387\u9ad8\u8fbe 79.8% (\u4fdd\u7559 97% \u6559\u5e08\u6027\u80fd!)]<\/p>\n<h3>\u4e00\u3001\u975e\u5bf9\u79f0\u8de8\u6a21\u6001\u84b8\u998f\u4e0e\u7a7a\u95f4\u6295\u5f71\u7684\u5fae\u89c2\u6570\u5b66\u63a8\u5bfc<\/h3>\n<p>\u8bbe\u6559\u5e08\u89c6\u89c9\u7f16\u7801\u5668\u4e3a $f_V^T(\\\\cdot)$&#xff0c;\u6559\u5e08\u6587\u672c\u7f16\u7801\u5668\u4e3a $f_T^T(\\\\cdot)$&#xff1b;\u5b66\u751f\u8f7b\u91cf\u7aef\u4fa7\u6587\u672c\u7f16\u7801\u5668\u4e3a $f_T^S(\\\\cdot)$\u3002\u56fe\u6587\u5bf9\u6279\u6b21\u4e3a ${(I_i, T_i)}_{i&#061;1}^B$\u3002<\/p>\n<h4>1. \u6559\u5e08\u7f51\u7edc\u7684\u9ec4\u91d1\u8de8\u6a21\u6001\u76f8\u4f3c\u5ea6\u8f6f\u6982\u7387\u5206\u5e03&#xff08;Soft Target Distribution&#xff09;<\/h4>\n<p>\u6587\u672c\u5230\u56fe\u50cf\u7684\u68c0\u7d22\u6982\u7387\u5206\u5e03&#xff1a;<\/p>\n<p>$$P_{i, j}^T &#061; \\\\frac{\\\\exp(\\\\langle f_T^T(T_i), f_V^T(I_j) \\\\rangle \/ \\\\tau_T)}{\\\\sum_{k&#061;1}^B \\\\exp(\\\\langle f_T^T(T_i), f_V^T(I_k) \\\\rangle \/ \\\\tau_T)}$$<\/p>\n<h4>2. \u975e\u5bf9\u79f0\u5b66\u751f\u76f8\u4f3c\u5ea6\u5206\u5e03&#xff08;Asymmetric Student Distribution&#xff09;<\/h4>\n<p>\u5b66\u751f\u6587\u672c\u6a21\u578b\u76f4\u63a5\u4e0e\u672a\u538b\u7f29\u7684\u5f3a\u5927\u6559\u5e08\u89c6\u89c9\u7279\u5f81\u8fdb\u884c\u5bf9\u9f50&#xff1a;<\/p>\n<p>$$P_{i, j}^S &#061; \\\\frac{\\\\exp(\\\\langle \\\\mathbf{W}{\\\\text{proj}} f_T^S(T_i), f_V^T(I_j) \\\\rangle \/ \\\\tau_S)}{\\\\sum{k&#061;1}^B \\\\exp(\\\\langle \\\\mathbf{W}_{\\\\text{proj}} f_T^S(T_i), f_V^T(I_k) \\\\rangle \/ \\\\tau_S)}$$<\/p>\n<h4>3. \u975e\u5bf9\u79f0\u8de8\u6a21\u6001 KL \u6563\u5ea6\u84b8\u998f\u635f\u5931&#xff1a;<\/h4>\n<p>$$\\\\mathcal{L}{\\\\text{asym_kd}} &#061; \\\\text{KL}\\\\left( P^T \\\\parallel P^S \\\\right) &#061; \\\\sum{i&#061;1}^B \\\\sum_{j&#061;1}^B P_{i, j}^T \\\\log \\\\left( \\\\frac{P_{i, j}^T}{P_{i, j}^S} \\\\right)$$<\/p>\n<ul>\n<li>\u4ee3\u6570\u4f18\u52bf&#xff1a;\u5b66\u751f\u6587\u672c\u7f51\u7edc\u88ab\u8feb\u5728\u6781\u5c0f\u7684 4 \u5c42\u53c2\u6570\u7a7a\u95f4\u5185&#xff0c;\u6700\u5927\u5316\u6a21\u62df\u5927\u6a21\u578b\u5728\u56fe\u6587\u7a7a\u95f4\u4e0a\u7684\u5168\u90e8\u7ec6\u7c92\u5ea6\u6392\u5e8f\u903b\u8f91&#xff0c;\u5f7b\u5e95\u6253\u7834\u4e86\u201c\u5b66\u751f\u5fc5\u987b\u4e0e\u540c\u7b49\u4f53\u91cf\u5b66\u751f\u5bf9\u9f50\u201d\u7684\u5bf9\u79f0\u6027\u80fd\u74f6\u9888&#xff01;<\/li>\n<\/ul>\n<h3>\u4e8c\u3001\u975e\u5bf9\u79f0\u56fe\u6587\u68c0\u7d22\u84b8\u998f\u4e0e\u91cf\u5316 PyTorch \u5de5\u4e1a\u7ea7\u5b9e\u73b0<\/h3>\n<p>import torch<br \/>\nimport torch.nn as nn<br \/>\nimport torch.nn.functional as F<br \/>\nfrom typing import Tuple<\/p>\n<p>class AsymmetricCrossModalDistillationEngine(nn.Module):<br \/>\n    &#034;&#034;&#034;<br \/>\n    \u975e\u5bf9\u79f0\u56fe\u6587\u68c0\u7d22\u84b8\u998f\u4e0e\u5bf9\u9f50\u5f15\u64ce<br \/>\n    &#034;&#034;&#034;<br \/>\n    def __init__(<br \/>\n        self,<br \/>\n        student_text_encoder: nn.Module,<br \/>\n        student_dim: int &#061; 256,<br \/>\n        teacher_dim: int &#061; 768,<br \/>\n        temperature: float &#061; 0.05<br \/>\n    ):<br \/>\n        super().__init__()<br \/>\n        self.student_text_encoder &#061; student_text_encoder<br \/>\n        # \u975e\u5bf9\u79f0\u7ef4\u5ea6\u6295\u5f71\u5bf9\u9f50\u5934<br \/>\n        self.proj_head &#061; nn.Linear(student_dim, teacher_dim, bias&#061;False)<br \/>\n        self.temperature &#061; temperature<\/p>\n<p>    def forward(<br \/>\n        self,<br \/>\n        text_tokens: torch.Tensor,<br \/>\n        teacher_image_embeds: torch.Tensor,<br \/>\n        teacher_text_embeds: torch.Tensor<br \/>\n    ) -&gt; Tuple[torch.Tensor, torch.Tensor]:<br \/>\n        &#034;&#034;&#034;<br \/>\n        teacher_image_embeds, teacher_text_embeds: \u9884\u5148\u63d0\u53d6\u5e76\u51bb\u7ed3\u7684\u9ad8\u7cbe\u6559\u5e08\u7279\u5f81 [B, 768]<br \/>\n        &#034;&#034;&#034;<br \/>\n        # 1. \u5b66\u751f\u8f7b\u91cf\u6587\u672c\u7f16\u7801\u524d\u5411<br \/>\n        raw_student_feat &#061; self.student_text_encoder(text_tokens)<br \/>\n        student_text_embeds &#061; F.normalize(self.proj_head(raw_student_feat), dim&#061;-1)<\/p>\n<p>        # 2. \u6559\u5e08\u9ec4\u91d1\u8de8\u6a21\u6001\u76f8\u4f3c\u5ea6\u77e9\u9635\u4e0e\u8f6f\u6982\u7387<br \/>\n        with torch.no_grad():<br \/>\n            t_sim &#061; torch.matmul(teacher_text_embeds, teacher_image_embeds.t()) \/ self.temperature<br \/>\n            t_prob &#061; F.softmax(t_sim, dim&#061;-1)<\/p>\n<p>        # 3. \u975e\u5bf9\u79f0\u5b66\u751f\u8de8\u6a21\u6001\u76f8\u4f3c\u5ea6\u77e9\u9635<br \/>\n        s_sim &#061; torch.matmul(student_text_embeds, teacher_image_embeds.t()) \/ self.temperature<br \/>\n        s_log_prob &#061; F.log_softmax(s_sim, dim&#061;-1)<\/p>\n<p>        # 4. KL \u6563\u5ea6\u84b8\u998f\u635f\u5931<br \/>\n        kd_loss &#061; F.kl_div(s_log_prob, t_prob, reduction&#061;&#039;batchmean&#039;)<\/p>\n<p>        # \u8f85\u4ee5 InfoNCE \u786c\u5bf9\u6bd4\u635f\u5931<br \/>\n        labels &#061; torch.arange(text_tokens.size(0), device&#061;text_tokens.device)<br \/>\n        hard_contrastive_loss &#061; F.cross_entropy(s_sim, labels)<\/p>\n<p>        total_loss &#061; 0.7 * kd_loss &#043; 0.3 * hard_contrastive_loss<br \/>\n        return total_loss, student_text_embeds<\/p>\n<h3>\u4e09\u3001\u771f\u5b9e\u79fb\u52a8\u7aef\u8fb9\u7f18\u82af\u7247&#xff08;\u9ad8\u901a\u9a81\u9f99 8 Gen 3 \/ Jetson Orin&#xff09;\u5b9e\u6d4b\u5bf9\u8d26<\/h3>\n<p>\u6211\u4eec\u5728\u57fa\u4e8e COCO \u56fe\u6587\u68c0\u7d22\u57fa\u51c6\u4e0e\u771f\u5b9e\u76f8\u518c\u4ee5\u56fe\u641c\u56fe\u6570\u636e\u96c6&#xff08;10 \u4e07\u5f20\u9ad8\u6e05\u56fe\u50cf\u5e93&#xff09;\u4e0a&#xff0c;\u5bf9\u6bd4\u4e86\u672a\u7ecf\u84b8\u998f\u7684\u7aef\u4fa7\u5c0f\u6a21\u578b\u4e0e\u975e\u5bf9\u79f0\u84b8\u998f\u6a21\u578b\u7684\u5b9e\u6d4b\u5bf9\u8d26&#xff1a;<\/p>\n<table>\n<tr>\u6a21\u578b\u67b6\u6784\u4e0e\u90e8\u7f72\u65b9\u6848\u7aef\u4fa7\u6587\u672c\u7f16\u7801\u5ef6\u8fdf (ms)\u6587\u672c\u6a21\u578b\u5185\u5b58\u5360\u7528 (MB)COCO Text-to-Image Top-1 \u53ec\u56de\u7387Top-5 \u7efc\u5408\u68c0\u7d22\u53ec\u56de\u7387<\/tr>\n<tbody>\n<tr>\n<td>\u539f\u59cb\u4e91\u7aef\u5927\u6a21\u578b (ViT-L\/14 FP16)<\/td>\n<td>45.0 ms (\u7aef\u4fa7\u6781\u6162\u6389\u5e27!)<\/td>\n<td>850 MB (\u5360\u7528\u8fc7\u5927)<\/td>\n<td>82.5% (\u6ee1\u8840\u5929\u82b1\u677f)<\/td>\n<td>96.8%<\/td>\n<\/tr>\n<tr>\n<td>\u7aef\u4fa7\u539f\u751f\u8bad\u7ec3\u8f7b\u91cf\u5c0f\u6a21\u578b (\u65e0\u84b8\u998f)<\/td>\n<td>1.8 ms (\u6781\u901f)<\/td>\n<td>32 MB<\/td>\n<td>38.4% (\u7cbe\u5ea6\u60e8\u4e0d\u5fcd\u7779!)<\/td>\n<td>62.0%<\/td>\n<\/tr>\n<tr>\n<td>\u4f20\u7edf\u5bf9\u79f0\u6a21\u578b\u84b8\u998f (\u5c0f\u6559\u5c0f)<\/td>\n<td>1.8 ms<\/td>\n<td>32 MB<\/td>\n<td>58.5%<\/td>\n<td>79.2%<\/td>\n<\/tr>\n<tr>\n<td>\u975e\u5bf9\u79f0\u5927\u8de8\u6a21\u6001\u84b8\u998f &#043; INT8 \u91cf\u5316<\/td>\n<td>1.8 ms (\u63d0\u901f 25x!)<\/td>\n<td>18 MB (\u6781\u5ea6\u7d27\u51d1!)<\/td>\n<td>79.8% (\u4fdd\u7559 96.7% \u6559\u5e08\u7cbe\u5ea6!)<\/td>\n<td>95.2% (\u8fd1\u4e4e\u7edd\u5bf9\u65e0\u635f!)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<h4>\u6838\u5fc3\u6536\u76ca\u5256\u6790&#xff1a;<\/h4>\n<li>\u7aef\u4fa7\u63a8\u7406\u8017\u65f6\u66b4\u964d 25 \u500d&#xff1a;\u4ece 45ms \u6781\u901f\u538b\u964d\u81f3 1.8 \u6beb\u79d2&#xff0c;\u5185\u5b58\u5f00\u9500\u4ec5\u9700 18MB&#xff0c;\u5728\u624b\u673a\u7aef\u4fa7\u5b9e\u73b0\u771f\u6b63\u7684\u201c\u5373\u8f93\u5373\u641c\u3001\u96f6\u5361\u987f\u54cd\u5e94\u201d&#xff1b;<\/li>\n<li>\u68c0\u7d22\u7cbe\u5ea6\u4fdd\u7559\u7387\u9ad8\u8fbe 96.7%&#xff1a;\u975e\u5bf9\u79f0\u8de8\u6a21\u6001\u84b8\u998f\u6210\u529f\u5c06\u5e9e\u5927\u6559\u5e08\u89c6\u89c9\u7a7a\u95f4\u7684\u7ec6\u7c92\u5ea6\u51e0\u4f55\u6d41\u5f62\u65e0\u7f1d\u704c\u6ce8\u8fdb 4 \u5c42\u5fae\u5c0f\u6587\u672c\u6a21\u578b\u4e2d&#xff0c;\u53ec\u56de\u7387\u7a81\u7834 79.8%&#xff01;<\/li>\n<h3>\u56db\u3001\u7ed3\u8bed<\/h3>\n<p>\u5728\u8fb9\u7f18\u667a\u80fd\u7684\u4e25\u82db\u7269\u7406\u73b0\u5b9e\u4e2d&#xff0c;\u7b97\u6cd5\u7684\u827a\u672f\u5728\u4e8e\u6253\u7834\u5bf9\u79f0\u7684\u6559\u6761&#xff0c;\u7528\u975e\u5bf9\u79f0\u7684\u7ed3\u6784\u91cd\u5851\u6548\u7387\u7684\u8fb9\u754c\u3002\u770b\u900f\u8de8\u6a21\u6001\u76f8\u4f3c\u5ea6\u6982\u7387\u5206\u5e03\u7684\u84b8\u998f\u672c\u8d28&#xff0c;\u7528\u6781\u7b80\u7684\u6587\u672c\u5b66\u751f\u5bf9\u9f50\u5f3a\u5927\u7684\u89c6\u89c9\u6559\u5e08&#xff0c;\u624d\u80fd\u8ba9\u524d\u6cbf\u7684\u591a\u6a21\u6001\u68c0\u7d22\u667a\u80fd\u5728\u65b9\u5bf8\u82af\u7247\u4e4b\u4e0a\u7efd\u653e\u51fa\u6700\u8000\u773c\u7684\u751f\u4ea7\u529b\u5149\u8292\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u5de5\u4e1a\u7ea7\u9ad8\u7cbe\u5ea6\u591a\u6a21\u6001\u56fe\u6587\u68c0\u7d22\u5728\u4f4e\u5ef6\u8fdf\u8fb9\u7f18\u8bbe\u5907\u4e0a\u7684\u91cf\u5316\u4e0e\u975e\u5bf9\u79f0\u8de8\u6a21\u6001\u84b8\u998f\u5b9e\u6218\u5728\u79fb\u52a8\u7aef\u76f8\u518c\u4ee5\u56fe\u641c\u56fe\u3001\u8f66\u8f7d\u591a\u6a21\u6001\u4eba\u673a\u4ea4\u4e92&#xff08;\u5982\u8bed\u97f3\u95ee\u7b54\u68c0\u7d22\u4e2d\u63a7\u5927\u5c4f\u89c6\u89c9\u76ee\u6807&#xff09;\u3001\u4ee5\u53ca\u624b\u6301\u5de1\u68c0\u7ec8\u7aef\u56fe\u6587\u67e5\u91cd\u7b49\u8fb9\u7f18\u7aef\u90e8\u7f72&#xff08;Edge Deployment&#xff09;\u4e2d&#xff0c;\u57fa\u4e8e CLIP \u67b6\u6784\u7684\u53cc\u5854\u56fe\u6587\u8de8\u6a21\u6001\u68c0\u7d22&#xff08;Dual-Tower Cross-modal Retrieval&#xff09; \u662f\u6838\u5fc3\u5e95\u5ea7\u3002<br \/>\n\u7136\u800c&#xff0c;\u5728\u9762\u5bf9**\u7b97\u529b\u6781<\/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":[74,66,2068,3677,50],"topic":[],"class_list":["post-110602","post","type-post","status-publish","format-standard","hentry","category-server","tag-agent","tag-ai","tag-nlp","tag-tensorflow","tag-50"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v20.3 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>\u5de5\u4e1a\u7ea7\u9ad8\u7cbe\u5ea6\u591a\u6a21\u6001\u56fe\u6587\u68c0\u7d22\u5728\u4f4e\u5ef6\u8fdf\u8fb9\u7f18\u8bbe\u5907\u4e0a\u7684\u91cf\u5316\u4e0e\u975e\u5bf9\u79f0\u8de8\u6a21\u6001\u84b8\u998f\u5b9e\u6218 - \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\/110602.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u5de5\u4e1a\u7ea7\u9ad8\u7cbe\u5ea6\u591a\u6a21\u6001\u56fe\u6587\u68c0\u7d22\u5728\u4f4e\u5ef6\u8fdf\u8fb9\u7f18\u8bbe\u5907\u4e0a\u7684\u91cf\u5316\u4e0e\u975e\u5bf9\u79f0\u8de8\u6a21\u6001\u84b8\u998f\u5b9e\u6218 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u5de5\u4e1a\u7ea7\u9ad8\u7cbe\u5ea6\u591a\u6a21\u6001\u56fe\u6587\u68c0\u7d22\u5728\u4f4e\u5ef6\u8fdf\u8fb9\u7f18\u8bbe\u5907\u4e0a\u7684\u91cf\u5316\u4e0e\u975e\u5bf9\u79f0\u8de8\u6a21\u6001\u84b8\u998f\u5b9e\u6218\u5728\u79fb\u52a8\u7aef\u76f8\u518c\u4ee5\u56fe\u641c\u56fe\u3001\u8f66\u8f7d\u591a\u6a21\u6001\u4eba\u673a\u4ea4\u4e92&#xff08;\u5982\u8bed\u97f3\u95ee\u7b54\u68c0\u7d22\u4e2d\u63a7\u5927\u5c4f\u89c6\u89c9\u76ee\u6807&#xff09;\u3001\u4ee5\u53ca\u624b\u6301\u5de1\u68c0\u7ec8\u7aef\u56fe\u6587\u67e5\u91cd\u7b49\u8fb9\u7f18\u7aef\u90e8\u7f72&#xff08;Edge Deployment&#xff09;\u4e2d&#xff0c;\u57fa\u4e8e CLIP \u67b6\u6784\u7684\u53cc\u5854\u56fe\u6587\u8de8\u6a21\u6001\u68c0\u7d22&#xff08;Dual-Tower Cross-modal Retrieval&#xff09; \u662f\u6838\u5fc3\u5e95\u5ea7\u3002 \u7136\u800c&#xff0c;\u5728\u9762\u5bf9**\u7b97\u529b\u6781\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/110602.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-09-28T19:48:49+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/09\/20260928194847-6abac49fb3e9a.webp\" \/>\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=\"3 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/110602.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/110602.html\",\"name\":\"\u5de5\u4e1a\u7ea7\u9ad8\u7cbe\u5ea6\u591a\u6a21\u6001\u56fe\u6587\u68c0\u7d22\u5728\u4f4e\u5ef6\u8fdf\u8fb9\u7f18\u8bbe\u5907\u4e0a\u7684\u91cf\u5316\u4e0e\u975e\u5bf9\u79f0\u8de8\u6a21\u6001\u84b8\u998f\u5b9e\u6218 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-09-28T19:48:49+00:00\",\"dateModified\":\"2026-09-28T19:48:49+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/110602.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/110602.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/110602.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u5de5\u4e1a\u7ea7\u9ad8\u7cbe\u5ea6\u591a\u6a21\u6001\u56fe\u6587\u68c0\u7d22\u5728\u4f4e\u5ef6\u8fdf\u8fb9\u7f18\u8bbe\u5907\u4e0a\u7684\u91cf\u5316\u4e0e\u975e\u5bf9\u79f0\u8de8\u6a21\u6001\u84b8\u998f\u5b9e\u6218\"}]},{\"@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":"\u5de5\u4e1a\u7ea7\u9ad8\u7cbe\u5ea6\u591a\u6a21\u6001\u56fe\u6587\u68c0\u7d22\u5728\u4f4e\u5ef6\u8fdf\u8fb9\u7f18\u8bbe\u5907\u4e0a\u7684\u91cf\u5316\u4e0e\u975e\u5bf9\u79f0\u8de8\u6a21\u6001\u84b8\u998f\u5b9e\u6218 - \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\/110602.html","og_locale":"zh_CN","og_type":"article","og_title":"\u5de5\u4e1a\u7ea7\u9ad8\u7cbe\u5ea6\u591a\u6a21\u6001\u56fe\u6587\u68c0\u7d22\u5728\u4f4e\u5ef6\u8fdf\u8fb9\u7f18\u8bbe\u5907\u4e0a\u7684\u91cf\u5316\u4e0e\u975e\u5bf9\u79f0\u8de8\u6a21\u6001\u84b8\u998f\u5b9e\u6218 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u5de5\u4e1a\u7ea7\u9ad8\u7cbe\u5ea6\u591a\u6a21\u6001\u56fe\u6587\u68c0\u7d22\u5728\u4f4e\u5ef6\u8fdf\u8fb9\u7f18\u8bbe\u5907\u4e0a\u7684\u91cf\u5316\u4e0e\u975e\u5bf9\u79f0\u8de8\u6a21\u6001\u84b8\u998f\u5b9e\u6218\u5728\u79fb\u52a8\u7aef\u76f8\u518c\u4ee5\u56fe\u641c\u56fe\u3001\u8f66\u8f7d\u591a\u6a21\u6001\u4eba\u673a\u4ea4\u4e92&#xff08;\u5982\u8bed\u97f3\u95ee\u7b54\u68c0\u7d22\u4e2d\u63a7\u5927\u5c4f\u89c6\u89c9\u76ee\u6807&#xff09;\u3001\u4ee5\u53ca\u624b\u6301\u5de1\u68c0\u7ec8\u7aef\u56fe\u6587\u67e5\u91cd\u7b49\u8fb9\u7f18\u7aef\u90e8\u7f72&#xff08;Edge Deployment&#xff09;\u4e2d&#xff0c;\u57fa\u4e8e CLIP \u67b6\u6784\u7684\u53cc\u5854\u56fe\u6587\u8de8\u6a21\u6001\u68c0\u7d22&#xff08;Dual-Tower Cross-modal Retrieval&#xff09; \u662f\u6838\u5fc3\u5e95\u5ea7\u3002 \u7136\u800c&#xff0c;\u5728\u9762\u5bf9**\u7b97\u529b\u6781","og_url":"https:\/\/www.wsisp.com\/helps\/110602.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-09-28T19:48:49+00:00","og_image":[{"url":"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/09\/20260928194847-6abac49fb3e9a.webp"}],"author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"3 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/110602.html","url":"https:\/\/www.wsisp.com\/helps\/110602.html","name":"\u5de5\u4e1a\u7ea7\u9ad8\u7cbe\u5ea6\u591a\u6a21\u6001\u56fe\u6587\u68c0\u7d22\u5728\u4f4e\u5ef6\u8fdf\u8fb9\u7f18\u8bbe\u5907\u4e0a\u7684\u91cf\u5316\u4e0e\u975e\u5bf9\u79f0\u8de8\u6a21\u6001\u84b8\u998f\u5b9e\u6218 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-09-28T19:48:49+00:00","dateModified":"2026-09-28T19:48:49+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/110602.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/110602.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/110602.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"\u5de5\u4e1a\u7ea7\u9ad8\u7cbe\u5ea6\u591a\u6a21\u6001\u56fe\u6587\u68c0\u7d22\u5728\u4f4e\u5ef6\u8fdf\u8fb9\u7f18\u8bbe\u5907\u4e0a\u7684\u91cf\u5316\u4e0e\u975e\u5bf9\u79f0\u8de8\u6a21\u6001\u84b8\u998f\u5b9e\u6218"}]},{"@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\/110602","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=110602"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/110602\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=110602"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=110602"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=110602"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=110602"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}