FSCD-147

最近引入了FSC-147数据集，用于在不同的147对象类别中具有6135个图像的少拍摄对象计数任务。在每个图像中，提供了所有对象的点注释和三个示例框。但是，此数据集不包含所有对象的边界框注释。出于评估目的，我们通过为val和测试集的所有对象提供边界框注释来扩展FSC-147数据集。我们命名新的数据集FSCD-147。

The recently proposed FSC-147 dataset, which contains 6,135 images spanning 147 distinct object categories, is designed for few-shot object counting tasks. For each image, point annotations for all objects and three example bounding boxes are provided, but the original dataset does not include bounding box annotations for all objects. For evaluation purposes, we extended the FSC-147 dataset by supplying bounding box annotations for all objects in its validation and test splits, and named the newly constructed dataset FSCD-147.

搜集汇总

背景与挑战

背景概述

FSCD-147是FSC-147数据集的扩展版本，增加了边界框注释，包含147个对象类别和6135张图像，用于少拍摄对象计数任务。该数据集由VinAI Research和石溪大学于2022年发布。