cy0307/ropedia-episode-task-suite-artifacts
收藏资源简介:
该数据集仓库包含由Ropedia发布的公开Xperience-10M样本episode的派生证据层,包括指标、预测、清单、图表、图表、笔记和重现脚本。它不包含原始Xperience-10M视频或原始annotation.hdf5文件。此仓库旨在提供项目的可审查部分,允许用户检查任务输出、比较提交的指标,并在不先下载原始视频的情况下理解单episode限制。包含的文件有results目录下的JSON和CSV文件(用于存储验证的指标、元数据、预测、混淆矩阵等)、docs/assets目录下的SVG和PNG图像(生成的图表、图表和ChatGPT图像支持的概述图)、docs/data/summary_metrics.json(仪表板可读的摘要包)、scripts目录下的Python重现脚本以及notes目录下的解释和可重现性笔记。该数据集验证了一个公共样本episode,具体包括5,821个对齐帧、1,161个滑动窗口、8,378个当前特征维度(音频已记录但未特征化)、12个监督/自监督任务定义,并按时间顺序分为70/30分割。
This dataset repo contains the derived evidence layer for the public Xperience-10M sample episode released by Ropedia: metrics, predictions, manifests, charts, diagrams, notes, and reproduction scripts. It does not contain raw Xperience-10M videos or raw annotation.hdf5. This repo is the reviewable half of the project, allowing users to inspect task outputs, compare committed metrics, and understand single-episode limitations without downloading raw videos first. Included files are JSON and CSV files in results/ (for verified metrics, metadata, predictions, confusion matrices, etc.), SVG and PNG images in docs/assets/ (generated diagrams, charts, and ChatGPT-image-backed overview figures), docs/data/summary_metrics.json (dashboard-readable summary bundle), Python reproduction scripts in scripts/, and interpretation and reproducibility notes in notes/. The artifacts validate one public sample episode with 5,821 aligned frames, 1,161 sliding windows, 8,378 current feature dimensions (audio documented but not featurized), 12 supervised/self-supervised task definitions, and a chronological 70/30 split.




