dataflow-cv 1.0.1

A computer vision dataset processing library

DataFlow-CV

Where Vibe Coding meets CV data. 🌊 Convert & visualize datasets. Built with the flow of Claude Code.

A computer vision dataset processing library for seamless format conversion and visualization between YOLO, LabelMe, and COCO annotation formats. Designed for researchers and developers working with multi-format annotation pipelines.

Features

• Bidirectional Conversion: Convert between YOLO, LabelMe, and COCO formats in any direction
• Multi-format Support: Handle object detection bounding boxes and instance segmentation polygons
• Native Coordinate Storage: Coordinates stored in format-native representation (YOLO normalized, LabelMe/COCO absolute pixels)
• Visualization: Visualize annotations with OpenCV, supporting both display and save modes
• Command-line Interface: User-friendly CLI with convert and visualize subcommands
• Python API: Programmatic access for integration into larger pipelines
• Verbose Logging: Detailed logging with file output for debugging
• Headless Mode: Run visualization in server/Docker environments with --no-display
• Flexible Error Handling: Choose between strict (abort on error) or lenient (skip and continue) modes
• Cross-platform: Full support for Windows, Linux, and macOS

Table of Contents

• DataFlow-CV
  • Features
  • Table of Contents
  • Installation
    • From PyPI
    • From Source
    • Optional Dependencies
  • Quick Start
    • Command-line Interface
      • Format Conversion
      • Visualization
    • Python API
  • Documentation
    • Key Concepts
  • Development
    • Testing
    • Linting and Formatting
    • Project Structure
  • Contributing
  • License
  • Acknowledgments

Installation

From PyPI

pip install dataflow-cv

From Source

# Clone the repository
git clone https://github.com/zjykzj/DataFlow-CV.git
cd DataFlow-CV

# Regular installation
pip install .

# Editable installation (for development)
pip install -e .

Note: When installed in editable mode, use python -m dataflow.cli instead of the dataflow-cv command.

Optional Dependencies

• pycocotools: Required for COCO RLE segmentation support

  pip install pycocotools
  

Quick Start

Command-line Interface

All required parameters (image directories, label directories, class files, output paths) are positional arguments for better usability. Use --help on any subcommand for detailed usage.

Format Conversion

# YOLO to COCO
dataflow-cv convert yolo2coco images/ yolo_labels/ classes.txt coco_annotations.json

# With RLE encoding
dataflow-cv convert yolo2coco images/ yolo_labels/ classes.txt coco_annotations.json --do-rle

# YOLO to LabelMe
dataflow-cv convert yolo2labelme images/ yolo_labels/ classes.txt labelme_json/

# LabelMe to YOLO
dataflow-cv convert labelme2yolo labelme_json/ classes.txt yolo_labels/

# LabelMe to COCO
dataflow-cv convert labelme2coco labelme_json/ classes.txt coco_annotations.json

# With RLE encoding
dataflow-cv convert labelme2coco labelme_json/ classes.txt coco_annotations.json --do-rle

# COCO to YOLO
dataflow-cv convert coco2yolo coco_annotations.json yolo_labels/

# COCO to LabelMe
dataflow-cv convert coco2labelme coco_annotations.json labelme_json/

# Enable verbose logging
dataflow-cv convert yolo2coco images/ yolo_labels/ classes.txt coco_annotations.json --verbose

# Disable strict mode (skip invalid annotations instead of aborting)
dataflow-cv convert yolo2coco --no-strict images/ yolo_labels/ classes.txt coco_annotations.json

Visualization

# Visualize YOLO annotations
dataflow-cv visualize yolo images/ yolo_labels/ classes.txt --save visualized/

# Visualize LabelMe annotations
dataflow-cv visualize labelme images/ labelme_json/ --save visualized/

# Visualize COCO annotations
dataflow-cv visualize coco images/ coco_annotations.json --save visualized/

# Enable verbose logging for detailed debug output
dataflow-cv visualize yolo --verbose images/ yolo_labels/ classes.txt --save visualized/

# Run on headless server (no display window)
dataflow-cv visualize yolo --no-display images/ yolo_labels/ classes.txt --save visualized/

Python API

from dataflow.convert import YoloAndCocoConverter
from dataflow.visualize import YOLOVisualizer

# Convert YOLO to COCO
converter = YoloAndCocoConverter(source_to_target=True, verbose=True, strict_mode=True)
result = converter.convert(
    source_path="yolo_labels/",
    target_path="coco_annotations.json",
    class_file="classes.txt",
    image_dir="images/",
    do_rle=False  # Set to True for RLE encoding
)

# Visualize YOLO annotations
visualizer = YOLOVisualizer(
    label_dir="yolo_labels/",
    image_dir="images/",
    class_file="classes.txt",
    is_show=True,
    is_save=True,
    output_dir="visualized/",
    verbose=True,
    strict_mode=True
)
result = visualizer.visualize()

See the samples/ directory for complete examples:

• samples/visualize/yolo_demo.py - YOLO visualization example
• samples/visualize/labelme_demo.py - LabelMe visualization example
• samples/visualize/coco_demo.py - COCO visualization example
• samples/convert/ - Conversion examples

Documentation

• CLAUDE.md: Detailed architecture, development guide, and known gotchas
• CHANGELOG.md: Version history and breaking changes
• specs/: Canonical specifications organized into two layers:
  • formats/ — External format contracts (YOLO, LabelMe, COCO) and conversion rules
  • modules/ — Internal module architecture, interface contracts, and dependency constraints

Key Concepts

• Format-Native Coordinates: Coordinates stored in each format's native representation — YOLO normalized [0,1] center-based, LabelMe/COCO absolute pixels top-left. See DatasetAnnotations.format to determine semantics
• Explicit Coordinate Transforms: Converters handle all coordinate transformations between formats. No hidden normalization — lossy vs lossless behavior is explicitly documented
• Strict Mode: Validation errors raise exceptions (default). Disable in CLI with --no-strict, or in Python API with strict_mode=False
• Verbose Logging: Detailed debug logs saved to files when --verbose is used. The CLI prints "Verbose log saved to:
• Headless Support: Use --no-display for servers/Docker; use --save to output visualization images without a window
• Keyboard Shortcuts: During visualization, press q or ESC to exit early; Enter/Space to advance; any other key continues
• Missing Image Handling: Missing images are skipped with warnings, allowing processing to continue
• RLE Mask Visualization: COCO RLE masks are displayed with semi-transparent fills for better visibility
• Color Management: Each class ID gets a unique color from an HSV-based palette for consistent visualization
• Specifications: The specs/ directory contains the canonical format and module specifications — the authoritative reference for expected behavior

Development

For detailed developer guidance including advanced test commands, debugging, and architecture overview, see CLAUDE.md.

Testing

289 tests, 73% code coverage.

# Run all tests
pytest

# Run tests with coverage report
pytest --cov=dataflow --cov-report=term

# Run specific test module
pytest tests/convert/test_yolo_and_coco.py

Coverage by module:

Module	Coverage	Notes
dataflow/label/	78%	Core data models (66%), handlers (67-82%)
dataflow/convert/	83%	Converters (81-92%), base pipeline (62%)
dataflow/visualize/	69%	Visualizers (94-100%), base drawing (79%)
dataflow/cli/	76%	CLI commands (47-96%), utils (86%)
dataflow/util/	87%	File ops (84%), logging (99%)

Linting and Formatting

# Install development dependencies
pip install -e .[dev]

# Format code
black dataflow tests samples

# Sort imports
isort dataflow tests samples

# Type checking
mypy dataflow

# Linting
flake8 dataflow tests samples

Pre-commit Hooks (Optional)

Automatically check code quality before each commit:

# Install pre-commit
pip install pre-commit

# Install git hooks (run once)
pre-commit install

# After this, every `git commit` will auto-run:
#   black (code formatting)
#   isort (import sorting)
#   flake8 (linting)
#   trailing-whitespace / end-of-file-fixer / check-yaml / check-toml

# Manually run against all files
pre-commit run --all-files

Project Structure

dataflow/
├── label/           # Annotation handlers + data models (YOLO, LabelMe, COCO)
├── convert/         # Format converters + RLE conversion utility
├── visualize/       # Visualization modules (OpenCV-based)
├── util/            # Logging and file operation utilities
└── cli/             # CLI entry point, commands, and validation
tests/               # Unit and integration tests (label, convert, visualize, cli, util)
samples/             # Python API usage examples (visualize, convert, label, cli)
assets/              # Test data organized by format (det/seg) and annotation type
specs/               # Canonical specifications (formats/ + modules/ layers)

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Before contributing, review CLAUDE.md for architecture and development patterns.

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add or update tests as needed
5. Ensure code passes formatting and linting checks
6. Submit a pull request

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Thanks to the creators of YOLO, LabelMe, and COCO formats for establishing these annotation standards
• Built with OpenCV, NumPy, and Click
• Inspired by the need for seamless format conversion in multi-tool CV pipelines