"What do saliency models predict?" Koehler, Guo, Zhang, Eckstein, Journal of Vision, 2014

Description of this data This data repository contains: 800 Images, observer fixations for three tasks (free search, saliency search, object search) for 19-22 subjects for the same 800 images. In addition, the data set includes explicit judgments of saliency for the 800 images (100 individuals). See paper for more details. https://jov.arvojournals.org/article.aspx?articleid=2121547&resultClick=1 If you use this data, please cite: "What do saliency models predict?" Koehler, Guo, Zhang, Eckstein, Journal of Vision, 2014, 14(2) DOI: 10.11 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Contact info: Katie Koehler ([e-mail]) Miguel Eckstein ([e-mail]) Contents of folder: Cues (folder): Images with the cue names for the object search task embedded into a mid-level grey background. Cued objects are absent in images 1-400 and present in images 401-800. _Left and _Right suffixes indicate the expected location of the object based on the context in the image (either on the left half or right half of the image). Odd numbered images contain expected object locations on the left half, even on right. Cues (text file): text file with all cues and image numbers listed on a separate line. Same as cues embedded in images in Cues folder (in case you prefer to generate the cue text yourself in an experiment). Images: 405x405px experimental images (used for all tasks). ObserverData.mat: Matlab file containing all experimental data (explicit judgments and eye movements) in the form of image coordinates. Variables in this file: explicit_judg: matrix containing click locations for each observer in the explicit judgment task. rows index different observers, columns index different images, 3rd dim=coords (x=1, y=2) freeview_x/y, objsearch_x/y, salview_x/y: eye movements locations for each eye movement task. Rows index different observers, columns index different images, 3rd dim indexes eye movement number within a single trial. x- and y-coordinates are recorded in separate matrices as labeled. All text files are labeled by task and contain the same information as the matlab file, along with observer number, image number, and fixation number (in the case of eye movement data). %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Notes about individual images: The resolution and content of each image varies. Depending on your purpose, the factors listed below may prompt you to exclude a few images from use. If you have recommendations for information to include about specific images, please notify the researchers (contact info listed above). Images with disrupted aspect ratios: Image 35 Image 64 Image 184 Image 427 Images with noticable artifacts/pixelation: Image 97 Image 108 Image 184 Image 195 Image 234 Image 251 Image 545 Images with highly sparse content (repetitive background with no clear standout region or foreground object): Image 55 Image 79 Image 92 Image 97 Image 144 Image 147 Image 153 Image 182 Image 232 Image 520

本数据集说明如下：
本数据仓库包含以下内容：800张实验图像，以及针对这800张相同图像、由19至22名被试完成三项任务——自由搜索（free search）、显著性搜索（saliency search）、目标搜索（object search）——时采集的观察者眼动轨迹数据。此外，本数据集还包含这800张图像的显性显著性判断（explicit judgments of saliency）数据，该数据由100名被试完成。更多细节请参阅论文：https://jov.arvojournals.org/article.aspx?articleid=2121547&resultClick=1。若使用本数据集，请引用以下文献："What do saliency models predict?" Koehler, Guo, Zhang, Eckstein, 《视觉（Journal of Vision）》, 2014, 14(2) DOI: 10.11

联系信息如下：Katie Koehler（邮箱：[e-mail]）；Miguel Eckstein（邮箱：[e-mail]）

文件夹内容说明：
1. Cues文件夹：存储带有任务提示名称的图像，提示名称嵌入在中灰色背景中。目标搜索任务中的提示对象在图像1-400中不存在，在401-800中存在。文件名后缀"_Left"与"_Right"表示基于图像上下文的预期目标位置：分别对应图像左半区与右半区。其中，奇数编号图像的预期目标位置位于左半区，偶数编号图像则位于右半区。
2. Cues文本文件：包含所有提示词与图像编号的文本文件，每行对应一组信息，与Cues文件夹中嵌入了提示词的图像内容完全一致。若您希望在实验中自行生成提示文本，该文件可提供便利。
3. Images文件夹：存储分辨率为405×405像素的实验图像，该图像集可用于全部三项任务。
4. ObserverData.mat：Matlab格式文件，以图像坐标形式存储所有实验数据（包括显性判断数据与眼动数据）。该文件内的变量说明如下：
- explicit_judg：存储显性判断任务中每名观察者的点击位置的矩阵。行索引对应不同观察者，列索引对应不同图像，第三维为坐标维度（x=1，y=2）。
- freeview_x/y、objsearch_x/y、salview_x/y：分别对应三项眼动任务中每次注视点的位置坐标。行索引对应不同观察者，列索引对应不同图像，第三维索引单次试次内的注视点序号。x与y坐标分别存储于对应标签的独立矩阵中。
所有文本文件均按任务类型命名，其包含的信息与Matlab文件一致，同时还额外包含观察者编号、图像编号以及注视点序号（针对眼动数据）。

单张图像使用注意事项：
每张图像的分辨率与内容均存在差异。根据您的实际使用需求，以下因素可能提示您排除部分图像：
1. 宽高比异常的图像：图像35、64、184、427
2. 存在明显伪影/像素化的图像：图像97、108、184、195、234、251、545
3. 内容高度稀疏的图像（背景重复且无明显突出区域或前景对象）：图像55、79、92、97、144、147、153、182、232、520

若您有关于特定图像的补充信息建议，请告知上述研究人员（联系方式见前文）。