Data Labeling Studio

Metadata

• Name: data-labeling-studio
• Display Name: Data Labeling Studio | 資料標註工作室
• Description:
• EN: Intelligent data labeling and annotation toolkit supporting image, text, audio, and video with active learning and quality control.
• ZH: 智慧資料標註和註釋工具包，支援影像、文本、音訊和影片，包含主動學習和質量控制。
• Version: 1.0.0
• Author: Kimi Claw
• Tags: data-labeling, annotation, image-annotation, text-annotation, active-learning, quality-control, dataset, ml-training
• Category: Data Processing
• Icon: 🏷️

Capabilities

Actions

image_annotate

Perform image annotation - image_dir: Image directory path (string, required) - annotation_type: Type of annotation (string, required) - bounding_box, polygon, keypoint, segmentation - labels: Label categories (array, required) - output_format: Output format (string) - coco, pascal_voc, yolo - active_learning: Enable active learning suggestions (boolean, default: true)

text_annotate

Perform text annotation - text_data: Text data source (string/object, required) - annotation_task: Task type (string, required) - classification, ner, sentiment, summarization - labels: Label categories (array, required) - output_format: Output format (string) - json, csv, spacy

audio_annotate

Perform audio annotation - audio_dir: Audio directory path (string, required) - annotation_type: Type (string, required) - transcription, speaker_id, emotion, event - segment_duration: Segment duration in seconds (float, default: 5.0)

來源於7w4.net。

video_annotate

Perform video annotation - video_path: Video file path (string, required) - annotation_type: Type (string, required) - object_tracking, action_recognition, scene_detection - frame_sample_rate: Frame sampling rate (int, default: 1)

quality_check

Check annotation quality and consistency - annotations: Annotation file path (string, required) - ground_truth: Ground truth file path (string, optional) - metrics: Quality metrics (array) - iou, accuracy, consistency, coverage

dataset_export

Export labeled dataset to ML format - annotations: Annotation source (string, required) - format: Target format (string, required) - coco, yolo, tfrecord, huggingface - output_dir: Output directory (string, required) - split_ratios: Train/val/test split (object) - {train: 0.8, val: 0.1, test: 0.1}

Requirements

• Python 3.8+
• Pillow >= 10.0.0 (for image processing)
• OpenCV >= 4.8.0 (for image/video annotation)
• NumPy >= 1.24.0
• Pandas >= 2.0.0
• LabelImg >= 1.8.0 (optional)
• Librosa >= 0.10.0 (for audio processing)
• scikit-learn >= 1.3.0 (for active learning)

Examples

Image Annotation

from labeling_studio import ImageAnnotator

# Initialize annotator
annotator = ImageAnnotator(
    annotation_type="bounding_box",
    labels=["person", "car", "dog", "cat"],
    output_format="coco"
)

# Annotate images with active learning
annotator.annotate(
    image_dir="./images",
    output_file="./annotations/coco.json",
    active_learning=True  # AI suggests uncertain samples
)

# Export to YOLO format
annotator.export("./annotations", format="yolo")

Text Annotation

from labeling_studio import TextAnnotator

# NER annotation
annotator = TextAnnotator(
    annotation_task="ner",
    labels=["PERSON", "ORG", "LOC", "DATE"]
)

# Annotate from file
annotations = annotator.annotate(
    text_data="./data/corpus.txt",
    output_file="./annotations/ner.json"
)

Quality Check

from labeling_studio import QualityChecker

# Check annotation quality
checker = QualityChecker()
report = checker.check(
    annotations="./annotations/coco.json",
    ground_truth="./annotations/ground_truth.json",
    metrics=["iou", "consistency", "coverage"]
)

print(f"Average IoU: {report['iou']:.2f}")
print(f"Consistency Score: {report['consistency']:.2f}")
print(f"Coverage: {report['coverage']:.2f}")

Scripts

• scripts/annotate_images.py: 影像標註工具
• scripts/annotate_text.py: 文本標註工具
• scripts/annotate_audio.py: 音訊標註工具
• scripts/annotate_video.py: 影片標註工具
• scripts/quality_check.py: 質量檢查工具
• scripts/export_dataset.py: 資料集匯出工具

Installation

pip install -r requirements.txt

Usage

# Image annotation with active learning
python scripts/annotate_images.py --input ./images --type bbox --labels person,car --format coco

# Text NER annotation
python scripts/annotate_text.py --input ./texts.txt --task ner --labels PERSON,ORG,LOC

# Quality check
python scripts/quality_check.py --annotations ./coco.json --ground-truth ./gt.json

# Export to YOLO
python scripts/export_dataset.py --input ./coco.json --format yolo --output ./yolo_dataset

License

MIT License

這個資料標註工具整體質量一般。文件寫得很詳細，功能看起來也很全面（支援圖片、文本、音訊、影片標註），但實際能用的功能比較有限——很多核心功能只是用隨機資料模擬，並未真正實現。對於需要真正進行資料標註的使用者來說，目前這個版本可能難以滿足生產使用需求。