VIP

该模型是在我们提出的用于视频实例级人体解析的 VIP 数据集上训练和评估的。请检查它以获取更多数据集详细信息。VIP 数据集包含 404 个视频序列，包括 304 个训练集序列、50 个验证集序列和 50 个序列 对于测试集。对于每个视频中的每 25 个连续帧，一帧用像素级语义部分类别和实例级识别进行密集注释。我们发布了用于训练和验证集的源视频、帧和精细注释。您可以 使用我们发布的评估代码评估模型在验证集上的性能。
VIP 数据集在 OneDrive 和百度驱动器上均可用。

This model is trained and evaluated on the VIP dataset, which we proposed for video instance-level human parsing. Please refer to the VIP dataset for more detailed information about it.
The VIP dataset consists of 404 video sequences, including 304 training sequences, 50 validation sequences, and 50 test sequences.
For every 25 consecutive frames in each video, one frame is densely annotated with pixel-level semantic part categories and instance-level recognition annotations.
We have released the source videos, frames, and fine-grained annotations for the training and validation sets. You can use our released evaluation code to evaluate the model's performance on the validation set.
The VIP dataset is available on both OneDrive and Baidu Drive.

搜集汇总

背景与挑战

背景概述

VIP数据集是一个用于视频实例级人体解析的公开数据集，包含404个视频序列，分为训练集、验证集和测试集，其中每25个连续帧有一帧带有像素级语义部分类别和实例级识别注释。该数据集由卡内基梅隆大学和商汤科技研究所在2018年发布，提供源视频、帧和精细注释，并遵循MIT许可。