Skip to content

Commit

Permalink
Site updated: 2024-08-24 23:06:33
Browse files Browse the repository at this point in the history
  • Loading branch information
AwayGithub committed Aug 24, 2024
1 parent 4ebe515 commit 6a54faf
Show file tree
Hide file tree
Showing 4 changed files with 327 additions and 14 deletions.
94 changes: 91 additions & 3 deletions 2024/08/21/SOR datasets conclude/index.html
Original file line number Diff line number Diff line change
Expand Up @@ -32,8 +32,9 @@
<meta property="og:image" content="https://vip.helloimg.com/i/2024/08/19/66c33c9544e81.png">
<meta property="og:image" content="https://vip.helloimg.com/i/2024/08/19/66c33c951d101.png">
<meta property="og:image" content="https://vip.helloimg.com/i/2024/08/19/66c304649eb4a.png">
<meta property="og:image" content="https://vip.helloimg.com/i/2024/08/24/66c9d87028bbf.png">
<meta property="article:published_time" content="2024-08-21T08:27:54.000Z">
<meta property="article:modified_time" content="2024-08-21T08:33:37.738Z">
<meta property="article:modified_time" content="2024-08-24T15:03:34.069Z">
<meta property="article:author" content="AwayX">
<meta property="article:tag" content="论文笔记">
<meta name="twitter:card" content="summary">
Expand Down Expand Up @@ -224,10 +225,16 @@ <h1 class="post-title" itemprop="name headline">
<i class="far fa-calendar"></i>
</span>
<span class="post-meta-item-text">发表于</span>


<time title="创建时间:2024-08-21 16:27:54 / 修改时间:16:33:37" itemprop="dateCreated datePublished" datetime="2024-08-21T16:27:54+08:00">2024-08-21</time>
<time title="创建时间:2024-08-21 16:27:54" itemprop="dateCreated datePublished" datetime="2024-08-21T16:27:54+08:00">2024-08-21</time>
</span>
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-calendar-check"></i>
</span>
<span class="post-meta-item-text">更新于</span>
<time title="修改时间:2024-08-24 23:03:34" itemprop="dateModified" datetime="2024-08-24T23:03:34+08:00">2024-08-24</time>
</span>
<span class="post-meta-item">
<span class="post-meta-item-icon">
<i class="far fa-folder"></i>
Expand Down Expand Up @@ -666,6 +673,87 @@ <h3 id="分割数据库"><a href="#分割数据库" class="headerlink" title="
<td>1. 尺寸1,024×2,048<br>2. 类别:Building Fence Other Pedestrian Pole RoadLine Road SideWalk Vegetation Vehicles Wall TrafficSign Sky Ground Bridge RailTrack GroundRai TrafficLight Static Dynamic Water Terrain<br>3. 使用CARLA模拟器创建,虚拟传感器套件由6个位于同一视点的针孔相机组成,以获得立方体全景图像,使用立方体贴图到等矩形投影算法将获得的立方体贴图全景重新投影到通用的等矩形格式中<br>4. 利用8个开源城市地图,并在每个地图中设置了100到120个初始收集点。我们的虚拟收集车辆按照模拟器的交通规则行驶。我们每50帧采样一次,并在每个初始收集点保留前10个关键帧图像。为了确保收集数据的多样性,我们调节了天气和时间条件</td>
<td></td>
</tr>
</tbody></table>
<h3 id="SOD-数据库"><a href="#SOD-数据库" class="headerlink" title="SOD 数据库"></a>SOD 数据库</h3><img src="https://vip.helloimg.com/i/2024/08/24/66c9d87028bbf.png" />

<blockquote>
<p>截至 2023 年</p>
</blockquote>
<table>
<thead>
<tr>
<th>数据库</th>
<th>使用的论文</th>
<th>创建的论文</th>
<th>规模</th>
<th>标注类型</th>
<th>建库过程</th>
<th>其它</th>
</tr>
</thead>
<tbody><tr>
<td>360SOD</td>
<td>Multi-Stage Salient Object Detection in 360° Omnidirectional Image Using Complementary Object-Level Semantic Information<br><strong>ETCI 2023</strong></td>
<td>Distortion-adaptive salient object detection in 360° omnidirectional images<br><strong>TVCG 2020</strong></td>
<td>500张(400张训练图像、100张测试图像)<br>分辨率512 × 1024</td>
<td>object-level、human fixation</td>
<td></td>
<td>第一个全景 SOD 数据集</td>
</tr>
<tr>
<td><strong>F-360iSOD</strong></td>
<td></td>
<td>A FIXATION-BASED 360◦ BENCHMARK DATASET FOR SALIENT OBJECT DETECTION<br><strong>ICIP 2020</strong></td>
<td>107张<br>1165个显著物体<br>分辨率512 × 1024</td>
<td>object+<strong>instance</strong> level</td>
<td></td>
<td><a target="_blank" rel="noopener" href="https://github.com/Jun-Pu/F-360iSOD">Jun-Pu&#x2F;F-360iSOD: A dataset for fixation-based salient object segmentation in 360° images (github.com)</a></td>
</tr>
<tr>
<td>360SSOD</td>
<td>Multi-Stage Salient Object Detection in 360° Omnidirectional Image Using Complementary Object-Level Semantic Information<br><strong>ETCI 2023</strong></td>
<td>Stage-wise salient object detection in 360° omnidirectional image via object-level semantical saliency ranking<br><strong>TVCG 2020</strong></td>
<td>1105张(850张训练图像、255张测试图像)<br>分辨率546 × 1024</td>
<td>object level</td>
<td></td>
<td><a target="_blank" rel="noopener" href="https://github.com/360-SSOD/download">360-SSOD&#x2F;download (github.com)</a></td>
</tr>
<tr>
<td>ASOD60K<br>(全景视频)<br>PAVS10K的前身</td>
<td>ASOD60K: An Audio-Induced Salient Object Detection Dataset for Panoramic Videos<br><strong>Arxiv 2021</strong></td>
<td>ASOD60K: An Audio-Induced Salient Object Detection Dataset for Panoramic Videos<br><strong>Arxiv 2021</strong></td>
<td>来自67个全景视频的62,455视频帧,其中10,465个关键帧被赋予了标签<br>分辨率4K<br></td>
<td>head movement (HM) and eye fixations, bounding boxes, object-level masks, and instance-level labels</td>
<td></td>
<td><a target="_blank" rel="noopener" href="https://github.com/PanoAsh/ASOD60K">https://github.com/PanoAsh/ASOD60K</a><br>视频具有超类和子类<br>花费一年建立数据集</td>
</tr>
<tr>
<td><strong>ODI-SOD</strong></td>
<td>View-Aware Salient Object Detection for 360∘ Omnidirectional Image<br><strong>TM 2022</strong></td>
<td>View-Aware Salient Object Detection for 360∘ Omnidirectional Image<br><strong>TM 2022</strong></td>
<td><strong>6263</strong>张分辨率不低于2K的RP图像<br>(从Flickr网站收集的1,151张图片和从YouTube精选的5,112帧视频)<br>2,000张图片的测试集<br>4,263张图片的训练集<br>分辨率不低于2K<br></td>
<td>object level</td>
<td>1. 使用不同的对象类别关键词(例如,人类、狗、建筑)在Flickr和YouTube上搜索全景资源,参考MS-COCO类别以涵盖各种真实世界场景。收集了8,896张图片和998个视频,包括不同的场景(例如,室内、室外)、不同的场合(例如,旅行、体育)、不同的运动模式(例如,移动、静态)和不同的视角。然后,所有视频都被采样成关键帧,并将不令人满意的图片或帧(例如,没有显著对象、质量低)剔除。<br>3. 首先,我们要求五位研究人员通过投票来判断对象的显著性,并选择显著的对象。其次,注释方面手动根据选定的显著对象标记二进制遮罩。最后,五位研究人员交叉检查二进制遮罩,以确保准确的像素级对象级注释。</td>
<td><a target="_blank" rel="noopener" href="https://github.com/iCVTEAM/ODI-SOD">iCVTEAM&#x2F;ODI-SOD: A 360° omnidirectional image-based salient object detection (SOD) dataset referred to as ODI-SOD with object-level pixel-wise annotation on equirectangular projection (ERP). (github.com)</a><br>所选图像的显着区域数量从一个到十个以上,显着区域的面积比从小于0.02%到大于65%,分辨率从2K到8K,一半以上的场景很复杂并且包含不同的对象</td>
</tr>
<tr>
<td><strong>PAVS10K</strong><br>(全景视频)</td>
<td>PAV-SOD: A New Task towards Panoramic Audiovisual Saliency Detection<br><strong>ACMMCC 2023</strong><br></td>
<td>PAV-SOD: A New Task towards Panoramic Audiovisual Saliency Detection<br><strong>ACMMCC 2023</strong></td>
<td>训练视频:40个,共5796帧<br><br>测试视频:27个共4669帧<br></td>
<td>instance level、眼动数据</td>
<td>1. 通过使用多个搜索关键词(例如,360°&#x2F;全景&#x2F;全向视频,空间音频,环境声学)从YouTube获取,涵盖了各种真实世界动态场景(例如,室内&#x2F;室外场景)、多种场合(例如,体育、旅行、音乐会、采访、戏剧)、不同的运动模式(例如,静态&#x2F;移动摄像机)以及多样化的对象类别(例如,人类、乐器、动物)<br>2. 获得了<strong>67个</strong>高质量的<strong>4K</strong>视频序列,手动将视频剪辑成小片段(平均29.6秒),以避免在收集眼动注视点时产生疲劳,总共有62,455帧,记录了62,455 × 40个<strong>眼动注视点</strong><br>3. 所有的视频片段都是通过内置有120 Hz采样率的<strong>Tobii眼动追踪器</strong><strong>HTC Vive</strong>头戴式显示器(HMD)来展示,并收集眼动注视点。观察者。我们招募了<strong>40名</strong>参与者(8名女性和32名男性),年龄在18到34岁之间,他们报告说视力正常或矫正到正常。20名参与者被随机选中观看<strong>单声道声音</strong>的视频(第一组),而其他参与者观看<strong>没有声音</strong>的视频(第二组)<br>4. 这67个子类别可以根据主要声源的线索被归类为三个超类别,即说话(例如,对话、独白)、音乐(例如,唱歌、演奏乐器)和杂项(例如,街道上汽车引擎和喇叭的声音、露天环境中的人群噪音)<br>5. 从总共62,455帧中以1&#x2F;6的采样率统一提取了<strong>10,465</strong>帧,用于像素级注释,使用<strong>CVAT工具箱</strong>进行手动标记<br>6. <strong>3位</strong>资深研究人员参与了基于注视的显著对象的10,465帧的手动注释,最终获得了19,904个实例级显著对象标签</td>
<td><a target="_blank" rel="noopener" href="https://github.com/ZHANG-Jun-Pu/PAV-SOD">https://github.com/ZHANG-Jun-Pu/PAV-SOD</a><br><strong>第一个</strong>用于全景视频SOD的数据集</td>
</tr>
<tr>
<td>未发布<br>(全景视频,<strong>SOR</strong>)</td>
<td>Instance-Level Panoramic Audio-Visual Saliency Detection and Ranking<br><strong>ACMMM 2024</strong></td>
<td>Instance-Level Panoramic Audio-Visual Saliency Detection and Ranking<br><strong>ACMMM 2024</strong></td>
<td></td>
<td>instance level</td>
<td>根据多个观察者的<strong>注意力转移</strong><strong>PAVS10K</strong>数据集提供了真实的显著性排名</td>
<td>未公开<br><strong>第一个</strong>用于全景视频SOR的数据集</td>
</tr>
</tbody></table>

</div>
Expand Down
Loading

0 comments on commit 6a54faf

Please sign in to comment.