jeffliulab/card-calibration-v1-data

--- license: mit language: en tags: - image - color-calibration - object-detection - yolo size_categories: - 1K<n<10K --- # Card Calibration Dataset (v1) 255 hand-collected photos of color calibration cards, with augmentations, YOLO annotations, and extracted feature crops. Used to train [`jeffliulab/card-calibration-v1`](https://huggingface.co/jeffliulab/card-calibration-v1). Live demo: [`jeffliulab/card-calibration-v1` Space](https://huggingface.co/spaces/jeffliulab/card-calibration-v1). ## Contents | Archive | Files | Description | |---|---|---| | `raw_photos.tar.gz` | 255 | Original hand-collected photos | | `augmented_images.tar.gz` | 2,294 | Albumentations-augmented variants (brightness/hue/blur/etc) | | `feature_crops.tar.gz` | 9,154 | Center-region patch crops (4 patches × ~2,294 images) | | `yolo_card_dataset.tar.gz` | 510 | YOLO stage-1 training set (card detection) | | `yolo_pattern_dataset.tar.gz` | 510 | YOLO stage-2 training set (pattern detection) | | `yolo_labeled_card.tar.gz` | 510 | Raw YOLO annotations (stage 1) | | `yolo_labeled_patterns.tar.gz` | 510 | Raw YOLO annotations (stage 2) | | `generalization_test.png` | 1 | Held-out generalization test photo | | `features/feature_0216.csv` | 1 | Final training feature CSV (12-D + target) | | `features/GridMean_0216.csv` | 1 | Intermediate grid mean values | | `features/unique_colors.csv` | 1 | Unique color reference values | ## Usage The companion repo provides a one-command download script: ```bash git clone https://github.com/jeffliulab/Color_Calibration.git cd Color_Calibration python scripts/dataset_download.py ``` This downloads all archives and extracts them into `data/` matching the layout expected by the training scripts in `src/`. ## Manual download ```python from huggingface_hub import snapshot_download snapshot_download( repo_id="jeffliulab/card-calibration-v1-data", repo_type="dataset", local_dir="./dataset_cache", ) ``` ## Known issues 6 augmented feature crop files are missing (gsutil transfer failures during the migration from GCS). These are non-critical center-region crops and do not affect training or evaluation. ## License MIT — both data and code are freely available for research and commercial use.许可证：MIT协议 语言：英语 标签： - 图像 - 色彩校准 - 目标检测 - YOLO 规模类别： - 1K<n<10K # 色彩校准卡数据集（v1版） 本数据集包含255张手动采集的色彩校准卡照片，附带数据增强结果、YOLO标注文件以及提取的特征裁剪图。 本数据集用于训练 [`jeffliulab/card-calibration-v1`](https://huggingface.co/jeffliulab/card-calibration-v1) 模型。在线演示可访问 [`jeffliulab/card-calibration-v1` 空间](https://huggingface.co/spaces/jeffliulab/card-calibration-v1)。 ## 数据集内容 | 归档文件 | 文件数量 | 说明 | |---|---|---| | `raw_photos.tar.gz` | 255 | 原始手动采集照片 | | `augmented_images.tar.gz` | 2,294 | 基于Albumentations库的数据增强图像变体（涵盖亮度、色相、模糊等变换） | | `feature_crops.tar.gz` | 9,154 | 中心区域补丁裁剪图（每张增强图像裁剪4个补丁，对应约2,294张增强图像） | | `yolo_card_dataset.tar.gz` | 510 | YOLO第一阶段训练集（用于卡片目标检测） | | `yolo_pattern_dataset.tar.gz` | 510 | YOLO第二阶段训练集（用于图案目标检测） | | `yolo_labeled_card.tar.gz` | 510 | 原始YOLO第一阶段标注文件 | | `yolo_labeled_patterns.tar.gz` | 510 | 原始YOLO第二阶段标注文件 | | `generalization_test.png` | 1 | 预留泛化测试照片 | | `features/feature_0216.csv` | 1 | 最终训练特征CSV文件（包含12维特征与目标标签） | | `features/GridMean_0216.csv` | 1 | 中间网格均值文件 | | `features/unique_colors.csv` | 1 | 唯一色彩参考值文件 | ## 使用方法 配套代码仓库提供了一键下载脚本： bash git clone https://github.com/jeffliulab/Color_Calibration.git cd Color_Calibration python scripts/dataset_download.py 该脚本将下载所有归档文件并解压至`data/`目录，其目录结构与`src/`下的训练脚本所需格式完全匹配。 ## 手动下载 您也可以通过以下Python代码手动下载数据集： python from huggingface_hub import snapshot_download snapshot_download( repo_id="jeffliulab/card-calibration-v1-data", repo_type="dataset", local_dir="./dataset_cache", ) ## 已知问题 在从谷歌云存储（GCS）迁移过程中，有6份增强特征裁剪图文件因gsutil传输失败丢失。此类文件为非关键的中心区域裁剪图，不会对模型训练与评估造成任何影响。 ## 许可证 MIT协议：本数据集与配套代码均可免费用于研究与商业用途。