<h1>Evaluation<aclass="headerlink"href="#evaluation"title="Permalink to this headline">¶</a></h1>
<divclass="section"id="image-reid">
<h2>Image ReID<aclass="headerlink"href="#image-reid"title="Permalink to this headline">¶</a></h2>
<ulclass="simple">
<li><strong>Market1501</strong>, <strong>DukeMTMC-reID</strong>, <strong>CUHK03 (767/700 split)</strong> and <strong>MSMT17</strong> have fixed split so keeping <codeclass="docutils literal notranslate"><spanclass="pre">split_id=0</span></code> is fine.</li>
<li><strong>CUHK03 (classic split)</strong> has 20 fixed splits, so do <codeclass="docutils literal notranslate"><spanclass="pre">split_id=0~19</span></code>.</li>
<li><strong>VIPeR</strong> contains 632 identities each with 2 images under two camera views. Evaluation should be done for 10 random splits. Each split randomly divides 632 identities to 316 train ids (632 images) and the other 316 test ids (632 images). Note that, in each random split, there are two sub-splits, one using camera-A as query and camera-B as gallery while the other one using camera-B as query and camera-A as gallery. Thus, there are totally 20 splits generated with <codeclass="docutils literal notranslate"><spanclass="pre">split_id</span></code> starting from 0 to 19. Models can be trained on <codeclass="docutils literal notranslate"><spanclass="pre">split_id=[0,</span><spanclass="pre">2,</span><spanclass="pre">4,</span><spanclass="pre">6,</span><spanclass="pre">8,</span><spanclass="pre">10,</span><spanclass="pre">12,</span><spanclass="pre">14,</span><spanclass="pre">16,</span><spanclass="pre">18]</span></code> (because <codeclass="docutils literal notranslate"><spanclass="pre">split_id=0</span></code> and <codeclass="docutils literal notranslate"><spanclass="pre">split_id=1</span></code> share the same train set, and so on and so forth.). At test time, models trained on <codeclass="docutils literal notranslate"><spanclass="pre">split_id=0</span></code> can be directly evaluated on <codeclass="docutils literal notranslate"><spanclass="pre">split_id=1</span></code>, models trained on <codeclass="docutils literal notranslate"><spanclass="pre">split_id=2</span></code> can be directly evaluated on <codeclass="docutils literal notranslate"><spanclass="pre">split_id=3</span></code>, and so on and so forth.</li>
<li><strong>CUHK01</strong> is similar to VIPeR in the split generation.</li>
<li><strong>GRID</strong> , <strong>PRID450S</strong>, <strong>iLIDS</strong> and <strong>PRID</strong> have 10 random splits, so evaluation should be done by varying <codeclass="docutils literal notranslate"><spanclass="pre">split_id</span></code> from 0 to 9.</li>
<li><strong>SenseReID</strong> has no training images and is used for evaluation only.</li>
</ul>
<divclass="admonition note">
<pclass="first admonition-title">Note</p>
<pclass="last">The <codeclass="docutils literal notranslate"><spanclass="pre">split_id</span></code> argument is defined in <codeclass="docutils literal notranslate"><spanclass="pre">ImageDataManager</span></code> and <codeclass="docutils literal notranslate"><spanclass="pre">VideoDataManager</span></code>. Please refer to <aclass="reference internal"href="pkg/data.html#torchreid-data"><spanclass="std std-ref">torchreid.data</span></a>.</p>
</div>
</div>
<divclass="section"id="video-reid">
<h2>Video ReID<aclass="headerlink"href="#video-reid"title="Permalink to this headline">¶</a></h2>
<ulclass="simple">
<li><strong>MARS</strong> and <strong>DukeMTMC-VideoReID</strong> have fixed single split so using <codeclass="docutils literal notranslate"><spanclass="pre">split_id=0</span></code> is ok.</li>
<li><strong>iLIDS-VID</strong> and <strong>PRID2011</strong> have 10 predefined splits so evaluation should be done by varying <codeclass="docutils literal notranslate"><spanclass="pre">split_id</span></code> from 0 to 9.</li>
Built with <ahref="http://sphinx-doc.org/">Sphinx</a> using a <ahref="https://github.com/rtfd/sphinx_rtd_theme">theme</a> provided by <ahref="https://readthedocs.org">Read the Docs</a>.