aditijc/smooth-hierarchy-ucf101-canary-v1
收藏Hugging Face2026-04-22 更新2026-04-26 收录
下载链接：
https://hf-mirror.com/datasets/aditijc/smooth-hierarchy-ucf101-canary-v1
下载链接
链接失效反馈官方服务：
资源简介：
smooth-hierarchy-ucf101 — Canary v1数据集是一个用于管道验证运行的实验数据集，主要验证FFmpeg、torchcodec和torch在NYU torch H200节点上的工作状态，以及4-loss FiLM U-Net的训练情况。数据集基于UCF101视频数据（101个类别，13,320个视频），图像大小为64×64，批量大小为128，最大训练周期为2。数据集包含多个检查点文件、作业脚本、日志文件、元数据文件和依赖锁定文件，用于确保实验的可重复性。
The smooth-hierarchy-ucf101 — Canary v1 dataset is an experimental dataset for pipeline-verification runs, not the main training run. It establishes the functionality of FFmpeg, torchcodec, and torch on NYU torch H200 nodes and the training of a 4-loss FiLM U-Net. The dataset is based on UCF101 video data (101 classes, 13,320 videos) with an image size of 64×64, batch size of 128, and max epochs of 2. It includes various checkpoint files, job scripts, log files, metadata files, and dependency lock files to ensure reproducibility of the experiment.
提供机构：
aditijc
