ATD-12K

ATD-12K 是一个大规模的动画三元组数据集，包括 12,000 个三元组（train10k，test2k），通过手动检查和具有丰富注释的 test2k，包括难度级别、运动的感兴趣区域（RoI）和运动标签类别 从多元化制片人制作的 30 系列电影（狂野和现代）中收集的数据集，总时长超过 25 小时。共有 101 个剪辑，两种分辨率（即 1920×1080、1280×720） 注意有些三联体有字幕、水印问题ATD-12K is a large-scale animated triplet dataset consisting of 12,000 total triplets, split into 10k training samples and 2k testing samples. The test2k subset has undergone manual verification and is equipped with rich annotations including difficulty level, region of interest (RoI) for motions, and motion label categories. This dataset is collected from 30 animated series produced by diverse creators, covering both wild and modern styles, with a total duration of over 25 hours. It includes 101 video clips with two resolution specifications: 1920×1080 and 1280×720. Note that some triplets have subtitle and watermark issues.

搜集汇总

背景与挑战

背景概述

ATD-12K是一个大规模动画三元组数据集，包含12,000个三元组，专门用于视频帧插值任务，具有丰富的测试集注释，包括难度级别和运动标签。数据来源于30部多元化电影，总时长超过25小时，提供两种分辨率，适用于动画视频处理研究。