PET

PET数据集由国立科学技术大学创建，专注于动物为中心的Pascal对象类别，包含4135张图像，涵盖鸟、猫、牛、狗、马和羊六种动物类别。数据集通过记录40名用户在自由观看和视觉搜索条件下的眼动数据，旨在研究不同任务对目标检测的影响。创建过程中，参与者需在两种任务条件下观看图像，每张图像观看时间为2秒。PET数据集的应用领域主要集中在通过分析眼动模式来提高动物类别的对象分类准确性，解决传统监督学习方法在处理动物类别时因形状和纹理变化大而难以准确检测的问题。The PET dataset was developed by the National University of Science and Technology, focusing on animal-centric Pascal object categories. It contains 4,135 images covering six animal categories: bird, cat, cow, dog, horse, and sheep. The dataset records eye-tracking data from 40 participants under two viewing conditions: free viewing and visual search, aiming to investigate the impact of different tasks on object detection. During the dataset creation process, participants were required to view each image for 2 seconds under both task conditions. The primary applications of the PET dataset center on improving the accuracy of animal category object classification by analyzing eye movement patterns, addressing the challenge that traditional supervised learning methods struggle to achieve accurate detection due to significant variations in shape and texture when processing animal categories.