segment-toolkit 1.0.1

A Python toolkit to convert between binary segmentation masks and YOLO labels

Segment Toolkit

A modern, robust, and premium Python package designed to bridge the gap between pixel-level binary segmentation masks and YOLO bounding box labels. It provides a bidirectional pipeline with exception handling, extensive logging, a command-line interface (CLI), and a Python API.

---

Features

• Bidirectional Conversion:
  • Forward Pipeline: Convert binary masks to YOLO format labels (supports standard axis-aligned or advanced minimum area rotated bounding boxes).
  • Reverse Pipeline: Reconstruct binary masks from YOLO labels.
• Automatic Dependency Installer: Missing required packages (numpy, opencv-python, pillow, pandas, matplotlib) are automatically detected and installed via pip upon package import or script execution.
• Robust Exception Handling: Try-catch blocks wrapped around file I/O, contour finding, and resizing to prevent application crashes on corrupted or missing files.
• Dynamic Dataset Matching: Read classification mappings (in CSV or JSON format) to automatically assign multi-class IDs matching standard dataset schemas (like the ISIC dataset).
• YOLO Dataset Splitting: Automatically shuffles and partitions images and labels into training and testing sets with customizable split ratios, creating standard data.yaml configs.
• Overlay Visualizer: Overlay bounding boxes and class indicators directly onto source images for annotation inspection.
• Dual Interface: Use as a command-line application (segment-toolkit) or import as a Python library (import segment_toolkit).

---

Installation

To install the toolkit locally in editable mode (missing dependencies will install automatically):

pip install -e .

Manual Installation

If you prefer to install dependencies manually before installing the toolkit:

pip install -r requirements.txt
pip install .

---

Usage

1. Command Line Interface (CLI)

The package installs a console script called segment-toolkit.

Convert Masks to YOLO Labels

• Single File Conversion:

  segment-toolkit mask-to-yolo \
    --image images/ISIC_0024310.jpg \
    --mask mask/ISIC_0024310_segmentation.png \
    --output-txt labels/ISIC_0024310.txt \
    --class-id 4
  

• Batch Directory Conversion:

  segment-toolkit mask-to-yolo \
    --image-dir images/ \
    --mask-dir mask/ \
    --output-dir labels/ \
    --ground-truth GroundTruth.csv
  

• Options:

  • --rotated: Use rotated minimum area rectangles (cv2.minAreaRect) instead of standard axis-aligned rectangles.
  • --resize WIDTH HEIGHT: Set target size for image and mask resizing (default: 640 640).

Convert YOLO Labels to Masks

• Single File Conversion:

  segment-toolkit yolo-to-mask \
    --label labels/ISIC_0024310.txt \
    --output-mask masks_reconstructed/ISIC_0024310_segmentation.png
  

• Batch Directory Conversion:

  segment-toolkit yolo-to-mask \
    --label-dir labels/ \
    --output-dir masks_reconstructed/
  

Visualize Bounding Boxes

Draw YOLO labels on top of the original image:

segment-toolkit visualize \
  --image images/ISIC_0024310.jpg \
  --label labels/ISIC_0024310.txt \
  --output visualization.png

Split Dataset

Organize folders into YOLO-compliant structure (dataset/train and dataset/test splits) and output data.yaml:

segment-toolkit split \
  --images images/ \
  --labels labels/ \
  --output dataset/ \
  --ratio 0.8 \
  --seed 42

---

Ground Truth Formats

The --ground-truth parameter in batch conversion supports both CSV and JSON formats.

CSV Format

Assumes the first column contains the image identifier/filename, and the subsequent columns represent binary indicator classes (where 1 indicates class presence). Example GroundTruth.csv:

image,MEL,NV,BCC,AKIEC,BKL,DF,VASC
ISIC_0024306,0,1,0,0,0,0,0
ISIC_0024310,1,0,0,0,0,0,0

JSON Format

Supports three distinct schemas:

1. Flat Dictionary (Format A): Maps image IDs directly to class integers or class name strings.

   {
     "ISIC_0024306": 5,
     "ISIC_0024310": "MEL"
   }
   

2. Nested Indicators (Format B): Maps image IDs to dictionaries of binary class indicators.

   {
     "ISIC_0024306": { "MEL": 0, "NV": 1, "BCC": 0 },
     "ISIC_0024310": { "MEL": 1, "NV": 0, "BCC": 0 }
   }
   

3. List of Records (Format C): A list of objects containing image IDs and class descriptors.

   [
     { "image": "ISIC_0024306", "class_id": 5 },
     { "image": "ISIC_0024310", "MEL": 1, "NV": 0 }
   ]
   

Note: Class name strings (like "MEL", "NV") are automatically mapped to standard ISIC IDs (AKIEC=0, BCC=1, BKL=2, DF=3, MEL=4, NV=5, VASC=6). Custom column names default to index-based IDs.

---

Python API

Import classes directly into your code to programmatically build custom pipelines:

from segment_toolkit import MaskToYoloConverter, YoloToMaskConverter

# 1. Convert mask to YOLO label
yolo_conv = MaskToYoloConverter(target_size=(640, 640), bbox_type="standard")
yolo_conv.convert_single(
    image_path="images/ISIC_0024310.jpg",
    mask_path="mask/ISIC_0024310_segmentation.png",
    output_txt_path="labels/ISIC_0024310.txt",
    class_id=4
)

# 2. Batch convert a folder of masks with a JSON ground truth
yolo_conv.convert_dataset(
    images_dir="images",
    masks_dir="mask",
    output_labels_dir="labels",
    ground_truth="GroundTruth.json"
)

---

Technical Details

Coordinate Conversion Math

Bounding Box Center Calculation (Pixel Space)

For standard bounding boxes, the pixel coordinates from boundingRect are (xmin, ymin, w_pixel, h_pixel). $$\text{Center } X \quad x_{center} = x_{min} + \frac{w_{pixel}}{2.0}$$ $$\text{Center } Y \quad y_{center} = y_{min} + \frac{h_{pixel}}{2.0}$$

Coordinate Normalization (YOLO Format)

All coordinates are normalized to the range [0.0, 1.0]: $$x_{norm} = \frac{x_{center}}{img_width}, \quad y_{norm} = \frac{y_{center}}{img_height}$$ $$w_{norm} = \frac{w_{pixel}}{img_width}, \quad h_{norm} = \frac{h_{pixel}}{img_height}$$

---

Author

Zakria Gamal

• Computer Vision and AI Engineer
• LinkedIn: Zakria Gamal