SNARE

该数据集名为SNARE基准，旨在应对语言定位任务，特别是基于比较性语言描述来解决对象参照物的识别问题。它包含了一系列相似的目标物体和干扰物体，用以挑战三维语言定位方法。数据集不仅包含视觉描述，还有遮蔽条件下的描述，每个物体都有8个不同的视角图像。该数据集被划分为训练集、验证集和测试集，其中验证集和测试集包含了未见过的物体类别。规模上，训练集包含6,153个实例，验证集有371个实例，测试集则有1,357个实例。该任务的目的是从自然语言描述中识别物体参照物。

The SNARE Benchmark is a dataset designed for language localization tasks, specifically addressing object referent recognition based on comparative natural language descriptions. It comprises a set of similar target objects and distractor objects to challenge 3D language localization methods. The dataset includes both visual descriptions and descriptions under occlusion conditions, with each object having 8 images captured from distinct viewpoints. This dataset is split into training, validation, and test sets, where the validation and test sets contain unseen object categories. In terms of scale, the training set contains 6,153 instances, the validation set has 371 instances, and the test set includes 1,357 instances. The goal of this task is to identify object referents from natural language descriptions.