The ACL-Synthetic

为了减轻在合成数据中产生常见本地化挑战的困难，我们通过设计收集了35个平均面积为40.9平方米的高质量室内场景，这些场景具有无纹理墙、重复枕头/图纸等，并提供了场景的火车/测试拆分 (火车/测试: 15/20场景)。我们将合成数据集命名为ACL-synthetic。To mitigate the common localization challenges inherent in synthetic data, we collected 35 high-quality indoor scenes with an average area of 40.9 square meters via targeted design. These scenes feature texture-less walls, repeated pillows/drawings, and the like, and are accompanied by a pre-defined training/test split (training: 15 scenes, test: 20 scenes). We named this synthetic dataset ACL-synthetic.

搜集汇总

背景与挑战

背景概述

该数据集为ACL-synthetic，是一个为缓解合成数据中常见局部化问题而构建的高质量室内场景合成数据集，包含35个平均面积40.9平方米的场景，并划分为15个训练场景和20个测试场景。