Mask R-CNN for Image Segmentation and Detection

This project implements Mask R-CNN using PyTorch for image segmentation and object detection. It includes custom datasets, training, evaluation, and logging. The code is flexible and supports training on partial datasets, checkpointing, and resuming training.

Table of Contents

• Mask R-CNN for Image Segmentation and Detection
  • Table of Contents
  • Features
  • Installation
  • Usage
    • Training
    • Testing
    • Resume training from a saved checkpoint
  • Configuration
  • Logging

Features

• Mask R-CNN implementation with PyTorch.
• Custom dataset support for PhenoBench and other datasets.
• Training, testing, and resuming training modes.
• Flexible argument-based configuration using argparse.
• Logging to console and file.
• Handles testing datasets with only images.

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/yourproject.git
   cd yourproject
   

2. Create a virtual environment and install dependencies:

    python -m venv venv
    source venv/bin/activate        # On Windows: venv\Scripts\activate
    pip install -r requirements.txt
   

Prepare your dataset in the datasets/PhenoBench/ directory:

train/ and val/ folders should include subfolders like images and leaf_instances. test/ folder only requires the images subfolder.

Usage

Run the main.py script with the desired mode and arguments. Use the –help flag to see all options:

python main.py --help

Training

Train a model from scratch:

python main.py --train \
  --train_path datasets/PhenoBench/train \
  --val_path datasets/PhenoBench/val \
  --num_epochs 10 \
  --batch_size_train 4 \
  --checkpoint checkpoints/model.pth

Testing

Evaluate a pretrained model:

python main.py --test \
  --test_path datasets/PhenoBench/test \
  --checkpoint checkpoints/model.pth

Resume training from a saved checkpoint

python main.py --resume \
  --train_path datasets/PhenoBench/train \
  --val_path datasets/PhenoBench/val \
  --num_epochs 5 \
  --checkpoint checkpoints/model.pth \
  --resume_checkpoint checkpoints/model_resumed.pth

Configuration

Key arguments for main.py:

--train_path, --val_path, --test_path: Paths to datasets.
--train_percent, --val_percent, --test_percent: Use only a percentage of data.
--num_epochs: Number of epochs for training.
--batch_size_train, --batch_size_test: Batch sizes for training and testing.
--checkpoint: Path to save/load checkpoints.
--log_level: Logging level (DEBUG, INFO, etc.).

Logging

Logs are saved in the results/logs/ directory by default. They include:

• Command-line arguments.
• Training and evaluation progress.
• Warnings and errors.
• Examples
• Train on a Small Dataset

python main.py --train \
  --train_path datasets/PhenoBench/train \
  --val_path datasets/PhenoBench/val \
  --train_percent 0.1 \
  --val_percent 0.1 \
  --num_epochs 1 \
  --batch_size_train 2 \
  --log_level DEBUG

Test on the Test Dataset

python main.py --test \
  --test_path datasets/PhenoBench/test \
  --checkpoint checkpoints/model.pth