index.html 31.2 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55
<!DOCTYPE html>
<!--[if IE 8]><html class="no-js lt-ie9" lang="en" > <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js" lang="en" > <!--<![endif]-->
<head>
  <meta charset="utf-8">
  <meta http-equiv="X-UA-Compatible" content="IE=edge">
  <meta name="viewport" content="width=device-width, initial-scale=1.0">
  
  
  <link rel="shortcut icon" href="../img/favicon.ico">
  <title>模型库 - PaddleSlim Docs</title>
  <link href='https://fonts.googleapis.com/css?family=Lato:400,700|Roboto+Slab:400,700|Inconsolata:400,700' rel='stylesheet' type='text/css'>

  <link rel="stylesheet" href="../css/theme.css" type="text/css" />
  <link rel="stylesheet" href="../css/theme_extra.css" type="text/css" />
  <link rel="stylesheet" href="//cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/styles/github.min.css">
  
  <script>
    // Current page data
    var mkdocs_page_name = "\u6a21\u578b\u5e93";
    var mkdocs_page_input_path = "model_zoo.md";
    var mkdocs_page_url = null;
  </script>
  
  <script src="../js/jquery-2.1.1.min.js" defer></script>
  <script src="../js/modernizr-2.8.3.min.js" defer></script>
  <script src="//cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/highlight.min.js"></script>
  <script>hljs.initHighlightingOnLoad();</script> 
  
</head>

<body class="wy-body-for-nav" role="document">

  <div class="wy-grid-for-nav">

    
    <nav data-toggle="wy-nav-shift" class="wy-nav-side stickynav">
      <div class="wy-side-nav-search">
        <a href=".." class="icon icon-home"> PaddleSlim Docs</a>
        <div role="search">
  <form id ="rtd-search-form" class="wy-form" action="../search.html" method="get">
    <input type="text" name="q" placeholder="Search docs" title="Type search term here" />
  </form>
</div>
      </div>

      <div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
	<ul class="current">
	  
          
            <li class="toctree-l1">
		
    <a class="" href="..">Home</a>
	    </li>
          
56 57 58 59 60
            <li class="toctree-l1 current">
		
    <a class="current" href="./">模型库</a>
    <ul class="subnav">
            
61
    <li class="toctree-l2"><a href="#1">1. 图像分类</a></li>
62 63 64 65 66
    
        <ul>
        
            <li><a class="toctree-l3" href="#11">1.1 量化</a></li>
        
67
            <li><a class="toctree-l3" href="#12">1.2 剪裁</a></li>
68 69 70 71 72 73 74 75 76 77 78 79
        
            <li><a class="toctree-l3" href="#13">1.3 蒸馏</a></li>
        
        </ul>
    

    <li class="toctree-l2"><a href="#2">2. 目标检测</a></li>
    
        <ul>
        
            <li><a class="toctree-l3" href="#21">2.1 量化</a></li>
        
80
            <li><a class="toctree-l3" href="#22">2.2 剪裁</a></li>
81 82 83 84 85 86 87 88 89 90 91 92
        
            <li><a class="toctree-l3" href="#23">2.3 蒸馏</a></li>
        
        </ul>
    

    <li class="toctree-l2"><a href="#3">3. 图像分割</a></li>
    
        <ul>
        
            <li><a class="toctree-l3" href="#31">3.1 量化</a></li>
        
93
            <li><a class="toctree-l3" href="#32">3.2 剪裁</a></li>
94 95 96 97 98 99 100
        
        </ul>
    

    </ul>
	    </li>
          
101 102
            <li class="toctree-l1">
		
103
    <span class="caption-text">教程</span>
104 105 106
    <ul class="subnav">
                <li class="">
                    
107
    <a class="" href="../tutorials/quant_post_demo/">离线量化</a>
108 109 110
                </li>
                <li class="">
                    
111
    <a class="" href="../tutorials/quant_aware_demo/">量化训练</a>
112 113 114
                </li>
                <li class="">
                    
115
    <a class="" href="../tutorials/quant_embedding_demo/">Embedding量化</a>
116 117 118
                </li>
                <li class="">
                    
119
    <a class="" href="../tutorials/nas_demo/">SA搜索</a>
120 121 122
                </li>
                <li class="">
                    
123
    <a class="" href="../tutorials/distillation_demo/">知识蒸馏</a>
124 125 126 127 128 129
                </li>
    </ul>
	    </li>
          
            <li class="toctree-l1">
		
130
    <span class="caption-text">API</span>
131 132 133
    <ul class="subnav">
                <li class="">
                    
134
    <a class="" href="../api/quantization_api/">量化</a>
135 136 137
                </li>
                <li class="">
                    
138
    <a class="" href="../api/prune_api/">剪枝与敏感度</a>
139 140 141
                </li>
                <li class="">
                    
142
    <a class="" href="../api/analysis_api/">模型分析</a>
143 144 145
                </li>
                <li class="">
                    
146
    <a class="" href="../api/single_distiller_api/">知识蒸馏</a>
147 148 149
                </li>
                <li class="">
                    
150 151 152 153
    <a class="" href="../api/nas_api/">SA搜索</a>
                </li>
                <li class="">
                    
154
    <a class="" href="../search_space/">搜索空间</a>
155 156 157 158
                </li>
                <li class="">
                    
    <a class="" href="../table_latency/">硬件延时评估表</a>
159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192
                </li>
    </ul>
	    </li>
          
            <li class="toctree-l1">
		
    <a class="" href="../algo/algo/">算法原理</a>
	    </li>
          
        </ul>
      </div>
      &nbsp;
    </nav>

    <section data-toggle="wy-nav-shift" class="wy-nav-content-wrap">

      
      <nav class="wy-nav-top" role="navigation" aria-label="top navigation">
        <i data-toggle="wy-nav-top" class="fa fa-bars"></i>
        <a href="..">PaddleSlim Docs</a>
      </nav>

      
      <div class="wy-nav-content">
        <div class="rst-content">
          <div role="navigation" aria-label="breadcrumbs navigation">
  <ul class="wy-breadcrumbs">
    <li><a href="..">Docs</a> &raquo;</li>
    
      
    
    <li>模型库</li>
    <li class="wy-breadcrumbs-aside">
      
193
        <a href="https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/docs/model_zoo.md"
194 195 196 197 198 199 200 201 202
          class="icon icon-github"> Edit on GitHub</a>
      
    </li>
  </ul>
  <hr/>
</div>
          <div role="main">
            <div class="section">
              
203
                <h2 id="1">1. 图像分类<a class="headerlink" href="#1" title="Permanent link">#</a></h2>
204
<p>数据集:ImageNet1000类</p>
205
<h3 id="11">1.1 量化<a class="headerlink" href="#11" title="Permanent link">#</a></h3>
206 207 208
<table>
<thead>
<tr>
209 210 211
<th align="center">模型</th>
<th align="center">压缩方法</th>
<th align="center">Top-1/Top-5 Acc</th>
212
<th align="center">模型体积(MB)</th>
213
<th align="center">下载</th>
214 215 216 217
</tr>
</thead>
<tbody>
<tr>
218
<td align="center">MobileNetV1</td>
219
<td align="center">FP32 baseline</td>
220 221 222
<td align="center">70.99%/89.68%</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
223 224
</tr>
<tr>
225 226 227 228 229
<td align="center">MobileNetV1</td>
<td align="center">quant_post</td>
<td align="center">xx%/xx%</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
230 231
</tr>
<tr>
232 233 234 235 236
<td align="center">MobileNetV1</td>
<td align="center">quant_aware</td>
<td align="center">xx%/xx%</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
237 238
</tr>
<tr>
239
<td align="center">MobileNetV2</td>
240
<td align="center">FP32 baseline</td>
241 242 243
<td align="center">72.15%/90.65%</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
244 245
</tr>
<tr>
246 247 248 249 250
<td align="center">MobileNetV2</td>
<td align="center">quant_post</td>
<td align="center">xx%/xx%</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
251 252
</tr>
<tr>
253
<td align="center">MobileNetV2</td>
254 255 256 257 258 259 260
<td align="center">quant_aware</td>
<td align="center">xx%/xx%</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
</tr>
<tr>
<td align="center">ResNet50</td>
261
<td align="center">FP32 baseline</td>
262 263 264
<td align="center">76.50%/93.00%</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
265 266
</tr>
<tr>
267
<td align="center">ResNet50</td>
268 269 270 271 272 273 274 275 276 277 278
<td align="center">quant_post</td>
<td align="center">xx%/xx%</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
</tr>
<tr>
<td align="center">ResNet50</td>
<td align="center">quant_aware</td>
<td align="center">xx%/xx%</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
279 280 281
</tr>
</tbody>
</table>
282
<h3 id="12">1.2 剪裁<a class="headerlink" href="#12" title="Permanent link">#</a></h3>
283 284 285
<table>
<thead>
<tr>
286 287 288
<th align="center">模型</th>
<th align="center">压缩方法</th>
<th align="center">Top-1/Top-5 Acc</th>
289
<th align="center">模型体积(MB)</th>
290
<th align="center">GFLOPs</th>
291
<th align="center">下载</th>
292 293 294 295
</tr>
</thead>
<tbody>
<tr>
296
<td align="center">MobileNetV1</td>
297
<td align="center">baseline</td>
298
<td align="center">70.99%/89.68%</td>
299 300
<td align="center">17</td>
<td align="center">1.11</td>
301
<td align="center"><a href="http://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV1_pretrained.tar">下载链接</a></td>
302 303
</tr>
<tr>
304
<td align="center">MobileNetV1</td>
305
<td align="center">uniform -50%</td>
306
<td align="center">69.4%/88.66% (-1.59%/-1.02%)</td>
307 308
<td align="center">9</td>
<td align="center">0.56</td>
309
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/MobileNetV1_uniform-50.tar">下载链接</a></td>
310 311
</tr>
<tr>
312
<td align="center">MobileNetV1</td>
313
<td align="center">sensitive -30%</td>
314
<td align="center">70.4%/89.3% (-0.59%/-0.38%)</td>
315 316
<td align="center">12</td>
<td align="center">0.74</td>
317
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/MobileNetV1_sensitive-30.tar">下载链接</a></td>
318 319
</tr>
<tr>
320 321
<td align="center">MobileNetV1</td>
<td align="center">sensitive -50%</td>
322
<td align="center">69.8% / 88.9% (-1.19%/-0.78%)</td>
323 324
<td align="center">9</td>
<td align="center">0.56</td>
325
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/MobileNetV1_sensitive-50.tar">下载链接</a></td>
326 327
</tr>
<tr>
328
<td align="center">MobileNetV2</td>
329
<td align="center">baseline</td>
330 331 332
<td align="center">72.15%/90.65%</td>
<td align="center">15</td>
<td align="center">0.59</td>
333
<td align="center"><a href="https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_pretrained.tar">下载链接</a></td>
334 335
</tr>
<tr>
336
<td align="center">MobileNetV2</td>
337
<td align="center">uniform -50%</td>
338
<td align="center">65.79%/86.11% (-6.35%/-4.47%)</td>
339 340
<td align="center">11</td>
<td align="center">0.296</td>
341
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/MobileNetV2_uniform-50.tar">下载链接</a></td>
342
</tr>
343
<tr>
344
<td align="center">ResNet34</td>
345
<td align="center">baseline</td>
346 347 348
<td align="center">72.15%/90.65%</td>
<td align="center">84</td>
<td align="center">7.36</td>
349
<td align="center"><a href="https://paddle-imagenet-models-name.bj.bcebos.com/ResNet34_pretrained.tar">下载链接</a></td>
350 351
</tr>
<tr>
352
<td align="center">ResNet34</td>
353
<td align="center">uniform -50%</td>
354
<td align="center">70.99%/89.95% (-1.36%/-0.87%)</td>
355 356
<td align="center">41</td>
<td align="center">3.67</td>
357
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/ResNet34_uniform-50.tar">下载链接</a></td>
358 359
</tr>
<tr>
360
<td align="center">ResNet34</td>
361
<td align="center">auto -55.05%</td>
362
<td align="center">70.24%/89.63% (-2.04%/-1.06%)</td>
363 364
<td align="center">33</td>
<td align="center">3.31</td>
365
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/ResNet34_auto-55.tar">下载链接</a></td>
366 367 368
</tr>
</tbody>
</table>
369
<h3 id="13">1.3 蒸馏<a class="headerlink" href="#13" title="Permanent link">#</a></h3>
370 371 372
<table>
<thead>
<tr>
373 374 375
<th align="center">模型</th>
<th align="center">压缩方法</th>
<th align="center">Top-1/Top-5 Acc</th>
376
<th align="center">模型体积(MB)</th>
377
<th align="center">下载</th>
378 379 380 381
</tr>
</thead>
<tbody>
<tr>
382
<td align="center">MobileNetV1</td>
383
<td align="center">student</td>
384 385
<td align="center">70.99%/89.68%</td>
<td align="center">17</td>
386
<td align="center"><a href="http://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV1_pretrained.tar">下载链接</a></td>
387 388
</tr>
<tr>
389
<td align="center">ResNet50_vd</td>
390
<td align="center">teacher</td>
391 392
<td align="center">79.12%/94.44%</td>
<td align="center">99</td>
393
<td align="center"><a href="https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_vd_pretrained.tar">下载链接</a></td>
394 395
</tr>
<tr>
396 397
<td align="center">MobileNetV1</td>
<td align="center">ResNet50_vd<sup><a href="#trans1">1</a></sup> distill</td>
398
<td align="center">72.77%/90.68% (+1.78%/+1.00%)</td>
399
<td align="center">17</td>
400
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/MobileNetV1_distilled.tar">下载链接</a></td>
401
</tr>
402
<tr>
403
<td align="center">MobileNetV2</td>
404
<td align="center">student</td>
405 406
<td align="center">72.15%/90.65%</td>
<td align="center">15</td>
407
<td align="center"><a href="https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV2_pretrained.tar">下载链接</a></td>
408 409
</tr>
<tr>
410 411
<td align="center">MobileNetV2</td>
<td align="center">ResNet50_vd distill</td>
412
<td align="center">74.28%/91.53% (+2.13%/+0.88%)</td>
413
<td align="center">15</td>
414
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/MobileNetV2_distilled.tar">下载链接</a></td>
415 416
</tr>
<tr>
417
<td align="center">ResNet50</td>
418
<td align="center">student</td>
419 420
<td align="center">76.50%/93.00%</td>
<td align="center">99</td>
421
<td align="center"><a href="http://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_pretrained.tar">下载链接</a></td>
422 423
</tr>
<tr>
424
<td align="center">ResNet101</td>
425
<td align="center">teacher</td>
426 427
<td align="center">77.56%/93.64%</td>
<td align="center">173</td>
428
<td align="center"><a href="http://paddle-imagenet-models-name.bj.bcebos.com/ResNet101_pretrained.tar">下载链接</a></td>
429 430 431 432
</tr>
<tr>
<td align="center">ResNet50</td>
<td align="center">ResNet101 distill</td>
433
<td align="center">77.29%/93.65% (+0.79%/+0.65%)</td>
434
<td align="center">99</td>
435
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/ResNet50_distilled.tar">下载链接</a></td>
436 437 438
</tr>
</tbody>
</table>
439 440 441 442 443 444 445
<div class="admonition note">
<p class="admonition-title">Note</p>
<p><a name="trans1">[1]</a>:带_vd后缀代表该预训练模型使用了Mixup,Mixup相关介绍参考<a href="https://arxiv.org/abs/1710.09412">mixup: Beyond Empirical Risk Minimization</a></p>
</div>
<h2 id="2">2. 目标检测<a class="headerlink" href="#2" title="Permanent link">#</a></h2>
<h3 id="21">2.1 量化<a class="headerlink" href="#21" title="Permanent link">#</a></h3>
<p>数据集: COCO 2017</p>
446 447 448
<table>
<thead>
<tr>
449 450 451 452 453 454 455
<th align="center">模型</th>
<th align="center">压缩方法</th>
<th align="center">数据集</th>
<th align="center">Image/GPU</th>
<th align="center">输入608 Box AP</th>
<th align="center">输入416 Box AP</th>
<th align="center">输入320 Box AP</th>
456
<th align="center">模型体积(MB)</th>
457
<th align="center">下载</th>
458 459 460 461
</tr>
</thead>
<tbody>
<tr>
462
<td align="center">MobileNet-V1-YOLOv3</td>
463
<td align="center">FP32 baseline</td>
464 465 466 467 468 469 470
<td align="center">COCO</td>
<td align="center">8</td>
<td align="center">29.3</td>
<td align="center">29.3</td>
<td align="center">27.1</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
471 472
</tr>
<tr>
473 474 475 476 477 478 479 480 481
<td align="center">MobileNet-V1-YOLOv3</td>
<td align="center">quant_post</td>
<td align="center">COCO</td>
<td align="center">8</td>
<td align="center">xx</td>
<td align="center">xx</td>
<td align="center">xx</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
482
</tr>
483
<tr>
484 485 486 487 488 489 490 491 492
<td align="center">MobileNet-V1-YOLOv3</td>
<td align="center">quant_aware</td>
<td align="center">COCO</td>
<td align="center">8</td>
<td align="center">xx</td>
<td align="center">xx</td>
<td align="center">xx</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
493 494
</tr>
<tr>
495
<td align="center">R50-dcn-YOLOv3 obj365_pretrain</td>
496
<td align="center">FP32 baseline</td>
497 498 499 500 501 502 503
<td align="center">COCO</td>
<td align="center">8</td>
<td align="center">41.4</td>
<td align="center">xx</td>
<td align="center">xx</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
504 505
</tr>
<tr>
506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525
<td align="center">R50-dcn-YOLOv3 obj365_pretrain</td>
<td align="center">quant_post</td>
<td align="center">COCO</td>
<td align="center">8</td>
<td align="center">xx</td>
<td align="center">xx</td>
<td align="center">xx</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
</tr>
<tr>
<td align="center">R50-dcn-YOLOv3 obj365_pretrain</td>
<td align="center">quant_aware</td>
<td align="center">COCO</td>
<td align="center">8</td>
<td align="center">xx</td>
<td align="center">xx</td>
<td align="center">xx</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
526 527 528
</tr>
</tbody>
</table>
529
<p>数据集:WIDER-FACE</p>
530 531 532
<table>
<thead>
<tr>
533 534 535 536 537
<th align="center">模型</th>
<th align="center">压缩方法</th>
<th align="center">Image/GPU</th>
<th align="center">输入尺寸</th>
<th align="center">Easy/Medium/Hard</th>
538
<th align="center">模型体积(MB)</th>
539
<th align="center">下载</th>
540 541 542 543
</tr>
</thead>
<tbody>
<tr>
544
<td align="center">BlazeFace</td>
545
<td align="center">FP32 baseline</td>
546 547 548 549 550
<td align="center">8</td>
<td align="center">640</td>
<td align="center">0.915/0.892/0.797</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
551 552
</tr>
<tr>
553 554 555 556 557 558 559
<td align="center">BlazeFace</td>
<td align="center">quant_post</td>
<td align="center">8</td>
<td align="center">640</td>
<td align="center">xx/xx/xx</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
560 561
</tr>
<tr>
562 563 564 565 566 567 568
<td align="center">BlazeFace</td>
<td align="center">quant_aware</td>
<td align="center">8</td>
<td align="center">640</td>
<td align="center">xx/xx/xx</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
569 570
</tr>
<tr>
571
<td align="center">BlazeFace-Lite</td>
572
<td align="center">FP32 baseline</td>
573 574 575 576 577
<td align="center">8</td>
<td align="center">640</td>
<td align="center">0.909/0.885/0.781</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
578 579
</tr>
<tr>
580 581 582 583 584 585 586
<td align="center">BlazeFace-Lite</td>
<td align="center">quant_post</td>
<td align="center">8</td>
<td align="center">640</td>
<td align="center">xx/xx/xx</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
587
</tr>
588
<tr>
589 590 591 592 593 594 595
<td align="center">BlazeFace-Lite</td>
<td align="center">quant_aware</td>
<td align="center">8</td>
<td align="center">640</td>
<td align="center">xx/xx/xx</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
596 597
</tr>
<tr>
598
<td align="center">BlazeFace-NAS</td>
599
<td align="center">FP32 baseline</td>
600 601 602
<td align="center">8</td>
<td align="center">640</td>
<td align="center">0.837/0.807/0.658</td>
603
<td align="center">xx</td>
604
<td align="center"><a href="">下载链接</a></td>
605 606
</tr>
<tr>
607 608 609 610 611
<td align="center">BlazeFace-NAS</td>
<td align="center">quant_post</td>
<td align="center">8</td>
<td align="center">640</td>
<td align="center">xx/xx/xx</td>
612
<td align="center">xx</td>
613
<td align="center"><a href="">下载链接</a></td>
614 615
</tr>
<tr>
616 617 618 619 620
<td align="center">BlazeFace-NAS</td>
<td align="center">quant_aware</td>
<td align="center">8</td>
<td align="center">640</td>
<td align="center">xx/xx/xx</td>
621
<td align="center">xx</td>
622
<td align="center"><a href="">下载链接</a></td>
623 624 625
</tr>
</tbody>
</table>
626
<h3 id="22">2.2 剪裁<a class="headerlink" href="#22" title="Permanent link">#</a></h3>
627
<p>数据集:Pasacl VOC &amp; COCO 2017</p>
628 629 630
<table>
<thead>
<tr>
631 632
<th align="center">模型</th>
<th align="center">压缩方法</th>
633
<th align="center">数据集</th>
634
<th align="center">Image/GPU</th>
635 636 637
<th align="center">输入608 Box AP</th>
<th align="center">输入416 Box AP</th>
<th align="center">输入320 Box AP</th>
638
<th align="center">模型体积(MB)</th>
639
<th align="center">GFLOPs (608*608)</th>
640
<th align="center">下载</th>
641 642 643 644 645
</tr>
</thead>
<tbody>
<tr>
<td align="center">MobileNet-V1-YOLOv3</td>
646
<td align="center">baseline</td>
647
<td align="center">Pascal VOC</td>
648
<td align="center">8</td>
649 650 651
<td align="center">76.2</td>
<td align="center">76.7</td>
<td align="center">75.3</td>
652 653 654
<td align="center">94</td>
<td align="center">40.49</td>
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/object_detection/yolov3_mobilenet_v1_voc.tar">下载链接</a></td>
655 656 657
</tr>
<tr>
<td align="center">MobileNet-V1-YOLOv3</td>
658 659
<td align="center">sensitive -52.88%</td>
<td align="center">Pascal VOC</td>
660
<td align="center">8</td>
661 662 663
<td align="center">77.6 (+1.4)</td>
<td align="center">77.7 (1.0)</td>
<td align="center">75.5 (+0.2)</td>
664 665 666
<td align="center">31</td>
<td align="center">19.08</td>
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/yolov3_mobilenet_v1_voc_prune.tar">下载链接</a></td>
667 668 669
</tr>
<tr>
<td align="center">MobileNet-V1-YOLOv3</td>
670
<td align="center">baseline</td>
671 672 673 674
<td align="center">COCO</td>
<td align="center">8</td>
<td align="center">29.3</td>
<td align="center">29.3</td>
675 676 677 678
<td align="center">27.0</td>
<td align="center">95</td>
<td align="center">41.35</td>
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/object_detection/yolov3_mobilenet_v1.tar">下载链接</a></td>
679 680 681
</tr>
<tr>
<td align="center">MobileNet-V1-YOLOv3</td>
682
<td align="center">sensitive -51.77%</td>
683
<td align="center">COCO</td>
684
<td align="center">8</td>
685 686 687
<td align="center">26.0 (-3.3)</td>
<td align="center">25.1 (-4.2)</td>
<td align="center">22.6 (-4.4)</td>
688 689 690
<td align="center">32</td>
<td align="center">19.94</td>
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/yolov3_mobilenet_v1_prune.tar">下载链接</a></td>
691 692
</tr>
<tr>
693
<td align="center">R50-dcn-YOLOv3</td>
694
<td align="center">baseline</td>
695
<td align="center">COCO</td>
696 697
<td align="center">8</td>
<td align="center">39.1</td>
698 699 700 701 702
<td align="center">-</td>
<td align="center">-</td>
<td align="center">177</td>
<td align="center">89.60</td>
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/object_detection/yolov3_r50vd_dcn.tar">下载链接</a></td>
703 704
</tr>
<tr>
705
<td align="center">R50-dcn-YOLOv3</td>
706
<td align="center">sensitive -9.37%</td>
707
<td align="center">COCO</td>
708
<td align="center">8</td>
709
<td align="center">39.3 (+0.2)</td>
710 711 712 713 714
<td align="center">-</td>
<td align="center">-</td>
<td align="center">150</td>
<td align="center">81.20</td>
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/yolov3_r50vd_dcn_prune.tar">下载链接</a></td>
715 716 717
</tr>
<tr>
<td align="center">R50-dcn-YOLOv3</td>
718
<td align="center">sensitive -24.68%</td>
719
<td align="center">COCO</td>
720
<td align="center">8</td>
721
<td align="center">37.3 (-1.8)</td>
722 723 724 725 726
<td align="center">-</td>
<td align="center">-</td>
<td align="center">113</td>
<td align="center">67.48</td>
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/yolov3_r50vd_dcn_prune578.tar">下载链接</a></td>
727 728
</tr>
<tr>
729
<td align="center">R50-dcn-YOLOv3 obj365_pretrain</td>
730
<td align="center">baseline</td>
731 732 733
<td align="center">COCO</td>
<td align="center">8</td>
<td align="center">41.4</td>
734 735 736 737 738
<td align="center">-</td>
<td align="center">-</td>
<td align="center">177</td>
<td align="center">89.60</td>
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/object_detection/yolov3_r50vd_dcn_obj365_pretrained_coco.tar">下载链接</a></td>
739 740 741
</tr>
<tr>
<td align="center">R50-dcn-YOLOv3 obj365_pretrain</td>
742
<td align="center">sensitive -9.37%</td>
743
<td align="center">COCO</td>
744
<td align="center">8</td>
745
<td align="center">40.5 (-0.9)</td>
746 747 748 749 750
<td align="center">-</td>
<td align="center">-</td>
<td align="center">150</td>
<td align="center">81.20</td>
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/yolov3_r50vd_dcn_obj365_pretrained_coco_prune.tar">下载链接</a></td>
751 752
</tr>
<tr>
753
<td align="center">R50-dcn-YOLOv3 obj365_pretrain</td>
754
<td align="center">sensitive -24.68%</td>
755
<td align="center">COCO</td>
756
<td align="center">8</td>
757
<td align="center">37.8 (-3.3)</td>
758 759 760 761 762
<td align="center">-</td>
<td align="center">-</td>
<td align="center">113</td>
<td align="center">67.48</td>
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/yolov3_r50vd_dcn_obj365_pretrained_coco_prune578.tar">下载链接</a></td>
763 764 765
</tr>
</tbody>
</table>
766 767
<h3 id="23">2.3 蒸馏<a class="headerlink" href="#23" title="Permanent link">#</a></h3>
<p>数据集:Pasacl VOC &amp; COCO 2017</p>
768 769 770
<table>
<thead>
<tr>
771 772 773 774 775 776 777
<th align="center">模型</th>
<th align="center">压缩方法</th>
<th align="center">数据集</th>
<th align="center">Image/GPU</th>
<th align="center">输入608 Box AP</th>
<th align="center">输入416 Box AP</th>
<th align="center">输入320 Box AP</th>
778
<th align="center">模型体积(MB)</th>
779
<th align="center">下载</th>
780 781 782 783
</tr>
</thead>
<tbody>
<tr>
784
<td align="center">MobileNet-V1-YOLOv3</td>
785
<td align="center">student</td>
786 787 788 789 790 791
<td align="center">Pascal VOC</td>
<td align="center">8</td>
<td align="center">76.2</td>
<td align="center">76.7</td>
<td align="center">75.3</td>
<td align="center">94</td>
792
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/object_detection/yolov3_mobilenet_v1_voc.tar">下载链接</a></td>
793 794
</tr>
<tr>
795
<td align="center">ResNet34-YOLOv3</td>
796
<td align="center">teacher</td>
797 798 799 800 801 802
<td align="center">Pascal VOC</td>
<td align="center">8</td>
<td align="center">82.6</td>
<td align="center">81.9</td>
<td align="center">80.1</td>
<td align="center">162</td>
803
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/object_detection/yolov3_r34_voc.tar">下载链接</a></td>
804 805
</tr>
<tr>
806 807 808 809
<td align="center">MobileNet-V1-YOLOv3</td>
<td align="center">ResNet34-YOLOv3 distill</td>
<td align="center">Pascal VOC</td>
<td align="center">8</td>
810 811 812
<td align="center">79.0 (+2.8)</td>
<td align="center">78.2 (+1.5)</td>
<td align="center">75.5 (+0.2)</td>
813
<td align="center">94</td>
814
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/yolov3_mobilenetv1_voc_distilled.tar">下载链接</a></td>
815
</tr>
816
<tr>
817
<td align="center">MobileNet-V1-YOLOv3</td>
818
<td align="center">student</td>
819 820 821 822 823 824
<td align="center">COCO</td>
<td align="center">8</td>
<td align="center">29.3</td>
<td align="center">29.3</td>
<td align="center">27.0</td>
<td align="center">95</td>
825
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/object_detection/yolov3_mobilenet_v1.tar">下载链接</a></td>
826 827
</tr>
<tr>
828
<td align="center">ResNet34-YOLOv3</td>
829
<td align="center">teacher</td>
830 831 832 833 834 835
<td align="center">COCO</td>
<td align="center">8</td>
<td align="center">36.2</td>
<td align="center">34.3</td>
<td align="center">31.4</td>
<td align="center">163</td>
836
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/object_detection/yolov3_r34.tar">下载链接</a></td>
837 838 839 840 841 842
</tr>
<tr>
<td align="center">MobileNet-V1-YOLOv3</td>
<td align="center">ResNet34-YOLOv3 distill</td>
<td align="center">COCO</td>
<td align="center">8</td>
843 844 845
<td align="center">31.4 (+2.1)</td>
<td align="center">30.0 (+0.7)</td>
<td align="center">27.1 (+0.1)</td>
846
<td align="center">95</td>
847
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/yolov3_mobilenetv1_coco_distilled.tar">下载链接</a></td>
848 849 850
</tr>
</tbody>
</table>
851 852 853
<h2 id="3">3. 图像分割<a class="headerlink" href="#3" title="Permanent link">#</a></h2>
<p>数据集:Cityscapes</p>
<h3 id="31">3.1 量化<a class="headerlink" href="#31" title="Permanent link">#</a></h3>
854 855 856
<table>
<thead>
<tr>
857 858 859
<th align="center">模型</th>
<th align="center">压缩方法</th>
<th align="center">mIoU</th>
860
<th align="center">模型体积(MB)</th>
861
<th align="center">下载</th>
862 863 864 865
</tr>
</thead>
<tbody>
<tr>
866
<td align="center">DeepLabv3+/MobileNetv1</td>
867
<td align="center">FP32 baseline</td>
868 869 870 871 872 873 874
<td align="center">63.26</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
</tr>
<tr>
<td align="center">DeepLabv3+/MobileNetv1</td>
<td align="center">quant_post</td>
875 876
<td align="center">xx</td>
<td align="center">xx</td>
877
<td align="center"><a href="">下载链接</a></td>
878
</tr>
879
<tr>
880 881 882 883 884
<td align="center">DeepLabv3+/MobileNetv1</td>
<td align="center">quant_aware</td>
<td align="center">xx</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
885 886
</tr>
<tr>
887
<td align="center">DeepLabv3+/MobileNetv2</td>
888
<td align="center">FP32 baseline</td>
889 890 891
<td align="center">69.81</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
892 893
</tr>
<tr>
894 895 896 897 898
<td align="center">DeepLabv3+/MobileNetv2</td>
<td align="center">quant_post</td>
<td align="center">xx</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
899 900
</tr>
<tr>
901 902 903 904 905
<td align="center">DeepLabv3+/MobileNetv2</td>
<td align="center">quant_aware</td>
<td align="center">xx</td>
<td align="center">xx</td>
<td align="center"><a href="">下载链接</a></td>
906 907 908
</tr>
</tbody>
</table>
909
<h3 id="32">3.2 剪裁<a class="headerlink" href="#32" title="Permanent link">#</a></h3>
910 911 912
<table>
<thead>
<tr>
913 914 915
<th align="center">模型</th>
<th align="center">压缩方法</th>
<th align="center">mIoU</th>
916
<th align="center">模型体积(MB)</th>
917
<th align="center">GFLOPs</th>
918
<th align="center">下载</th>
919 920 921 922
</tr>
</thead>
<tbody>
<tr>
923 924 925 926 927
<td align="center">fast-scnn</td>
<td align="center">baseline</td>
<td align="center">69.64</td>
<td align="center">11</td>
<td align="center">14.41</td>
928
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/fast_scnn_cityscape.tar">下载链接</a></td>
929 930
</tr>
<tr>
931 932 933 934 935
<td align="center">fast-scnn</td>
<td align="center">uniform  -17.07%</td>
<td align="center">69.58 (-0.06)</td>
<td align="center">8.5</td>
<td align="center">11.95</td>
936
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/fast_scnn_cityscape_uniform-17.tar">下载链接</a></td>
937 938 939 940 941 942 943
</tr>
<tr>
<td align="center">fast-scnn</td>
<td align="center">sensitive -47.60%</td>
<td align="center">66.68 (-2.96)</td>
<td align="center">5.7</td>
<td align="center">7.55</td>
944
<td align="center"><a href="https://paddlemodels.bj.bcebos.com/PaddleSlim/fast_scnn_cityscape_sensitive-47.tar">下载链接</a></td>
945
</tr>
946 947 948 949 950 951 952 953 954
</tbody>
</table>
              
            </div>
          </div>
          <footer>
  
    <div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
      
955
        <a href="../tutorials/quant_post_demo/" class="btn btn-neutral float-right" title="离线量化">Next <span class="icon icon-circle-arrow-right"></span></a>
956
      
957
      
958
        <a href=".." class="btn btn-neutral" title="Home"><span class="icon icon-circle-arrow-left"></span> Previous</a>
959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985
      
    </div>
  

  <hr/>

  <div role="contentinfo">
    <!-- Copyright etc -->
    
  </div>

  Built with <a href="http://www.mkdocs.org">MkDocs</a> using a <a href="https://github.com/snide/sphinx_rtd_theme">theme</a> provided by <a href="https://readthedocs.org">Read the Docs</a>.
</footer>
      
        </div>
      </div>

    </section>

  </div>

  <div class="rst-versions" role="note" style="cursor: pointer">
    <span class="rst-current-version" data-toggle="rst-current-version">
      
          <a href="https://github.com/PaddlePaddle/PaddleSlim/" class="fa fa-github" style="float: left; color: #fcfcfc"> GitHub</a>
      
      
986
        <span><a href=".." style="color: #fcfcfc;">&laquo; Previous</a></span>
987 988
      
      
989
        <span style="margin-left: 15px"><a href="../tutorials/quant_post_demo/" style="color: #fcfcfc">Next &raquo;</a></span>
990
      
991 992 993 994 995 996 997 998 999 1000
    </span>
</div>
    <script>var base_url = '..';</script>
    <script src="../js/theme.js" defer></script>
      <script src="../mathjax-config.js" defer></script>
      <script src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.0/MathJax.js?config=TeX-AMS-MML_HTMLorMML" defer></script>
      <script src="../search/main.js" defer></script>

</body>
</html>