Preprocessing/Cleaning/Labeling: No preprocessing or cleaning was done on the data before annotation. The raw data is provided along with the annotations.
Software Availability: None of the software used for preprocessing/cleaning/labeling the data is available.
Additional Dataset Details: The dataset is made up of 16,758 images with natural language descriptions, collected through a variety of sources including Amazon Mechanical Turk, Flickr, and Google Image Search. The descriptions are annotated with objects, scenes, and actions, and the dataset is divided into training, validation, and test sets.
Has an analysis of the potential impact of the dataset and its use on data subjects been conducted? No. All annotations are on objective world states with no subjective opinions or arguments involved.
Any other comments? N/A
Computer Science, Computer Vision and Pattern Recognition
Encoder Performance Comparison for Object Detection
