Datasets:
Add training materials manifest (462 videos / 7852 clips)
Browse files- README.md +101 -0
- summary.json +33 -0
- training-materials.csv +0 -0
- training-materials.txt +0 -0
README.md
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-nc-4.0
|
| 3 |
+
language:
|
| 4 |
+
- zh
|
| 5 |
+
- en
|
| 6 |
+
- ja
|
| 7 |
+
pretty_name: Easy4K SFDP · Siamese NET AI — Training Materials
|
| 8 |
+
task_categories:
|
| 9 |
+
- image-classification
|
| 10 |
+
tags:
|
| 11 |
+
- video
|
| 12 |
+
- frame-deduplication
|
| 13 |
+
- siamese-network
|
| 14 |
+
- knowledge-distillation
|
| 15 |
+
- data-provenance
|
| 16 |
+
size_categories:
|
| 17 |
+
- n<1K
|
| 18 |
+
---
|
| 19 |
+
|
| 20 |
+
# Easy4K SFDP · Siamese NET AI — 训练素材清单
|
| 21 |
+
|
| 22 |
+
本仓库是 [Easy4K SFDP · Siamese NET AI](https://huggingface.co/Sudabaka/SFDP-Siamese-NET-AI)
|
| 23 |
+
的**数据溯源清单(Data Manifest)**,记录模型训练所用视频的来源、作者、许可证与切片规模。
|
| 24 |
+
|
| 25 |
+
> **本仓库不包含任何视频文件。** 仅发布元数据,供版权核查、学术复现与审计使用。
|
| 26 |
+
|
| 27 |
+
## 数据集概述
|
| 28 |
+
|
| 29 |
+
| 项目 | 数值 |
|
| 30 |
+
|---|---|
|
| 31 |
+
| 素材视频 | 462 支 |
|
| 32 |
+
| 切片 | 7852 个 |
|
| 33 |
+
| 训练 / 验证样本 | 472,011 / 78,147(合计 550,158) |
|
| 34 |
+
| 素材类型 | 2D 动画(术力口 / MV)、真人 MV、Live、实拍街景、开源动画 |
|
| 35 |
+
| 标注方式 | Easy4K SFDP 四阶判决器(像素差 / dHash / QR / Farnebäck 光流)教师自动生成软标签 |
|
| 36 |
+
| 划分策略 | 按来源视频隔离,同一条视频不会同时出现在训练集与验证集 |
|
| 37 |
+
|
| 38 |
+
## 素材构成
|
| 39 |
+
|
| 40 |
+
| 来源 | 视频数 | 切片数 | 许可 / 说明 |
|
| 41 |
+
|---|---|---|---|
|
| 42 |
+
| Wikimedia Commons | 29 | 676 | CC BY / CC BY-SA / CC0,逐条标注作者与许可证 |
|
| 43 |
+
| 用户提供 MV | 10 | 288 | 由模型作者提供并授权训练 |
|
| 44 |
+
| Bilibili 公开 MV | 341 | 5640 | 仅训练使用,不随模型分发 |
|
| 45 |
+
| Bilibili 补充素材 | 78 | 1248 | 切片阶段补采,元数据后补,保留 bvid |
|
| 46 |
+
| 固定测试集 | 4 | 0 | 全新 MV,完全不参与训练 |
|
| 47 |
+
|
| 48 |
+
## 文件说明
|
| 49 |
+
|
| 50 |
+
| 文件 | 说明 |
|
| 51 |
+
|---|---|
|
| 52 |
+
| `training-materials.txt` | 人类可读完整清单,按来源分段,含标题 / 作者 / 许可 / 来源链接 / 切片数 |
|
| 53 |
+
| `training-materials.csv` | 机器可读版本 |
|
| 54 |
+
| `summary.json` | 汇总统计 |
|
| 55 |
+
|
| 56 |
+
### CSV 字段
|
| 57 |
+
|
| 58 |
+
| 字段 | 含义 |
|
| 59 |
+
|---|---|
|
| 60 |
+
| `section` | `wikimedia_commons` / `user_provided` / `bilibili` / `bilibili_supplement` / `heldout_test` |
|
| 61 |
+
| `file` | 训练时使用的本地文件名 |
|
| 62 |
+
| `title` | 原始标题 |
|
| 63 |
+
| `author` | 作者 / 上传者 |
|
| 64 |
+
| `license` | 许可证或使用声明 |
|
| 65 |
+
| `resolution` | 原始分辨率(CC 素材) |
|
| 66 |
+
| `duration_sec` | 时长(秒) |
|
| 67 |
+
| `quality` | 下载清晰度代码(Bilibili qn) |
|
| 68 |
+
| `source_url` | 原始页面链接 |
|
| 69 |
+
| `clips` | 该视频切出的片段数(按磁盘实际目录统计) |
|
| 70 |
+
|
| 71 |
+
## 溯源与合规
|
| 72 |
+
|
| 73 |
+
- **CC 素材**:逐条保留作者、许可证与来源页链接,使用方式符合 CC BY / CC BY-SA / CC0 条款;
|
| 74 |
+
本清单与模型卡共同构成署名(Attribution)。
|
| 75 |
+
- **Bilibili 素材**:仅用于本地训练,**不随模型分发、不提供下载**。清单保留 bvid 与作者,
|
| 76 |
+
便于权利人核查。权利人提出异议时,将从训练集中移除并在后续版本重训。
|
| 77 |
+
- **用户提供素材**:由模型作者本人提供并授权用于训练。
|
| 78 |
+
- **固定测试集**:与训练集完全隔离,避免指标虚高。
|
| 79 |
+
|
| 80 |
+
## 已知偏差
|
| 81 |
+
|
| 82 |
+
- 实拍街景与开源动画占比偏高,番剧 1 拍 2 / 1 拍 3 的真实样本相对有限;
|
| 83 |
+
- 素材以 1080p 及以上为主,低分辨率与强压缩内容覆盖不足;
|
| 84 |
+
- 分布以中文 / 日文音乐视频与动画为主,其他语种与题材覆盖有限。
|
| 85 |
+
|
| 86 |
+
## 引用
|
| 87 |
+
|
| 88 |
+
```bibtex
|
| 89 |
+
@misc{easy4k_sfdp_siamese_net_ai,
|
| 90 |
+
title = {Easy4K SFDP · Siamese NET AI},
|
| 91 |
+
author = {Sudabaka},
|
| 92 |
+
year = {2026},
|
| 93 |
+
howpublished = {\url{https://huggingface.co/Sudabaka/SFDP-Siamese-NET-AI}},
|
| 94 |
+
note = {Lightweight Siamese network for video frame deduplication}
|
| 95 |
+
}
|
| 96 |
+
```
|
| 97 |
+
|
| 98 |
+
## 许可证
|
| 99 |
+
|
| 100 |
+
本清单采用 [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/) 发布。
|
| 101 |
+
清单中各视频的版权归原作者所有,本仓库不授予任何视频内容的使用许可。
|
summary.json
ADDED
|
@@ -0,0 +1,33 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"generator": "Easy4K SFDP Siamese NET AI materials manifest",
|
| 3 |
+
"generated_at": "2026-10-18T09:36:34",
|
| 4 |
+
"total_videos": 462,
|
| 5 |
+
"total_clips": 7852,
|
| 6 |
+
"samples": {
|
| 7 |
+
"train": 472011,
|
| 8 |
+
"val": 78147,
|
| 9 |
+
"total": 550158
|
| 10 |
+
},
|
| 11 |
+
"by_section": {
|
| 12 |
+
"wikimedia_commons": {
|
| 13 |
+
"videos": 29,
|
| 14 |
+
"clips": 676
|
| 15 |
+
},
|
| 16 |
+
"user_provided": {
|
| 17 |
+
"videos": 10,
|
| 18 |
+
"clips": 288
|
| 19 |
+
},
|
| 20 |
+
"bilibili": {
|
| 21 |
+
"videos": 341,
|
| 22 |
+
"clips": 5640
|
| 23 |
+
},
|
| 24 |
+
"bilibili_supplement": {
|
| 25 |
+
"videos": 78,
|
| 26 |
+
"clips": 1248
|
| 27 |
+
},
|
| 28 |
+
"heldout_test": {
|
| 29 |
+
"videos": 4,
|
| 30 |
+
"clips": 0
|
| 31 |
+
}
|
| 32 |
+
}
|
| 33 |
+
}
|
training-materials.csv
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
training-materials.txt
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|