Pebble Contours

A contour segmentation tool for pebbles using YOLO-based AI segmentation. Includes an interactive Gradio web UI and batch-processing mode.

Table of Contents

• Prerequisites
• Installation
• Usage
  • Interactive Mode (Gradio)
  • Batch Inference
• Extra Scripts
• Example Interface
• License

Prerequisites

Make sure the following are installed on your system:

• Python 3.10
• Anaconda Distribution

To verify installation:

python3 -m pip --version
conda --version

Installation

1. Create a conda environment:

conda create -n pebble_contours python=3.10

2. Activate the environment and install dependencies:

conda activate pebble_contours
pip install -r requirements.txt

Usage

Interactive Mode (Gradio)

To launch the interactive Gradio interface:

python3 code/serving_gradio/gradio_server.py

You’ll get a local URL (e.g. http://127.0.0.1:7860) and possibly a temporary public Gradio link.

Features:

• Upload image
• Adjust:
  • Confidence threshold
  • Minimum area factor
  • Inference device (auto / cpu / gpu)
• Download processed image and CSV with:
  • Mask_ID
  • Centroid_X / Centroid_Y
  • Polygon_Coordinates

Batch Inference

To process all images in a folder:

python3 code/model_usage/final_inference_version_on_folder.py path/input/folder path/output/folder --confidence 0.6 --area_factor 0.15 --device cpu

Output:

• JPG with visualized masks
• CSV with mask ID, centroid, polygon coordinates

Extra Scripts

Located in the code/ directory.

code/dataset_creation/

• check_dataset.py: View YOLO annotations
• sv_checker.py: Visualize a single image
• sv_dataset_annotator.py: Auto-annotate with SAM
• prueba_inicial.py: OpenCV-only segmentation

code/model_training/

• yolotrain.py: YOLO training script
• smooth_dataset.py: Smooths contour labels
• select_frames.py: Filters valid frames from dataset

code/model_usage/

• final_inference_version_on_folder.py: Final batch inference
• inference_v1.py, inference_v2_*.py: Older versions

code/serving_gradio/

• gradio_server.py: UI web server
• inference_yolo_adapter.py: YOLO integration

License

This project is delivered as-is. For redistribution or commercial use, contact the author or contracting organization.