yolodt 0.2.0

YOLO Dataset Tools - Comprehensive toolkit for YOLO format dataset processing

🎯 YDT - YOLO Dataset Tools

A Professional Toolkit for YOLO Dataset Processing

English | 简体中文

---

✨ Features

🖼️ Image Processing 📐 SAHI-powered smart slicing 🔄 Rotation with OBB transformation 📏 Multi-method resize (scale & crop) 📍 Coordinate-based precision cropping 🎨 HSV color augmentation 🎬 Video frame extraction 🚀 Multi-threaded video processing	📊 Dataset Operations ✂️ Smart train/val split 🔗 Multi-dataset merging 🎲 Synthetic data generation 🤖 YOLO auto-labeling 📦 Batch processing
👁️ Visualization 🖼️ Interactive dataset browser ⌨️ Keyboard controls (n/p/q) 🎯 Class filtering 📸 Letterbox preview 🎨 Augmentation effects	🎯 Format Support 📐 OBB (Oriented Bounding Box) 📦 Standard BBox 🔄 Auto format detection ✨ Seamless conversion

🚀 Quick Start

# Clone and install
git clone https://github.com/yourusername/ydt.git
cd ydt/package
uv venv && uv pip install -e .

# Try it out!
ydt image slice -i ./images -o ./output
ydt image slice -i ./images -o ./output -c 3 -d 2
ydt viz dataset -i ./dataset

💻 Usage

Command Line

# Image processing
ydt image slice -i ./imgs -o ./out -c 3
ydt image slice -i ./imgs -o ./out -c 3 -d 2 -r 0.1 --overlap-vertical 0.05
ydt image resize -i ./images -o ./resized -s 640 800 1024
ydt image augment -i data.yaml -o ./aug
ydt image video -i ./videos -o ./frames -s 30
ydt image video -i ./videos -o ./frames --parallel -w 4
ydt image crop-coords -i ./images -o ./cropped -c "100,50,600,400"

# Dataset operations
ydt dataset split -i data.yaml -o ./split -r 0.8
ydt dataset merge -i ./ds1 ./ds2 -o ./merged
ydt dataset synthesize -t ./targets -b ./backgrounds -o ./synthetic --objects-per-image 2-5 --split train --train-ratio 0.8
ydt dataset auto-label -i ./images -m ./yolo11n.pt --format bbox -o ./labeled

# Visualization
ydt viz dataset -i ./dataset
ydt viz letterbox -i ./image.jpg
ydt viz augment -i ./image.jpg

Python API

from ydt.image import (
    slice_dataset,
    augment_dataset,
    extract_frames,
    process_images_multi_method,
    concat_images_horizontally,
    concat_images_vertically
)
from ydt.dataset import split_dataset, DatasetSynthesizer, auto_label_dataset
from ydt.visual import visualize_dataset, visualize_letterbox

# Slice large images
slice_dataset("./dataset", "./sliced", horizontal_count=3)

# Grid slicing (2×3 = 6 slices)
slice_dataset("./dataset", "./sliced", horizontal_count=2, vertical_count=3)

# Resize images with both scale and crop methods
process_images_multi_method("./images", "./resized", target_sizes=[640, 800, 1024])

# Concatenate images
concat_images_horizontally("img1.jpg", "img2.jpg", "output.jpg", alignment="center")

# Extract frames from videos
extract_frames("./videos", "./frames", step=30)

# Split dataset
split_dataset("./data.yaml", "./split", train_ratio=0.8)

# Auto-label images
result = auto_label_dataset(
    input_dir="./images",
    model_path="./yolo11n.pt",
    format_type="bbox",
    output_dir="./labeled"
)

# Generate synthetic dataset with custom parameters
synthesizer = DatasetSynthesizer(
    target_dir="./targets",
    background_dir="./backgrounds",
    output_dir="./synthetic",
    objects_per_image=(2, 5),  # 2-5 objects per image
    split_mode="trainval",      # Generate train+val
    train_ratio=0.8            # 80% train, 20% val
)
stats = synthesizer.synthesize_dataset(num_images=1000)

# Visualize dataset
visualize_dataset("./dataset", scan_train=True)
visualize_letterbox("./image.jpg", output_dir="./output")

📦 What's Inside

ydt/
├── 🖼️  image/       # Image processing
├── 📊  dataset/     # Dataset operations
├── 👁️  visual/      # Visualization
├── 🛠️  core/        # Core utilities
├── 🤖  auto_label/  # Auto-labeling
└── ⚡  cli/         # CLI interface

🎯 Key Features

Dual Format Support

Automatically detects and handles both formats:

Format	Values	Description
OBB	9 values	class_id x1 y1 x2 y2 x3 y3 x4 y4
BBox	5 values	class_id x_center y_center width height

Smart Slicing

Powered by SAHI, intelligently slice large images while preserving label accuracy. Supports both horizontal and grid slicing with configurable overlap ratios.

# Horizontal slicing (default)
ydt image slice -i ./images -o ./sliced -c 3 -r 0.1

# Grid slicing (3×2 = 6 slices)
ydt image slice -i ./images -o ./sliced -c 3 -d 2 -r 0.1 --overlap-vertical 0.05

# Fine grid slicing with custom overlap
ydt image slice -i ./images -o ./sliced -c 4 -d 3 -r 0.05 --overlap-vertical 0.02

Video Frame Extraction

Extract frames from video files for dataset creation. Supports both sequential and parallel processing.

# Sequential processing (default)
ydt image video -i ./videos -o ./frames -s 30

# Parallel processing for multiple videos
ydt image video -i ./videos -o ./frames --parallel -w 4

Features:

• 🎯 Smart worker count auto-detection
• ⚡ Concurrent video decoding
• 📊 Progress tracking per video
• 🔄 Automatic fallback for single videos

Auto-Labeling

Automatically label images using YOLO models with support for both BBox and OBB formats:

ydt dataset auto-label -i ./images -m ./yolo11n.pt --format bbox -o ./labeled

Features:

• 🎯 Support for both BBox and OBB formats
• 🤖 Automatic format detection and conversion
• 📁 Clean output directory structure
• ⚙️ Configurable confidence and IOU thresholds
• 🔍 Preview mode with --dry-run

Interactive Visualization

Browse your dataset with keyboard controls:

• n - Next image
• p - Previous image
• q - Quit

ydt viz dataset -i ./dataset

📝 License

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

🙏 Acknowledgments

• Ultralytics - YOLO framework
• SAHI - Slicing aided hyper inference
• Albumentations - Image augmentation

---