/dropletcode

Primary LanguageJupyter Notebook

Mask R-CNN Image Segmentation

This project performs instance segmentation on images using a pre-trained Mask R-CNN model.

Usage

The main steps are:

  1. Clone the Mask R-CNN repository
  2. Setup the Mask R-CNN model and load pre-trained weights
  3. Load and pre-process an input image
  4. Make predictions using the model
  5. Post-process and visualize results

Setup

  • Install dependencies: 
    pip install -r requirements.txt
  • Clone Mask R-CNN repo: 
    clone_repo("https://github.com/matterport/Mask_RCNN")
  • Setup model and load weights: 
    config = MaskRCNNConfig()
model = setup_model(config)
load_weights(model, weights_path)

Making Predictions

  • Load input image: 
    img = load_and_display(image_path)
  • Make predictions: 
    results = model.detect([img])
  • Visualize predictions: 
    display_predictions(img, results[0], class_names)

Post-processing

The results dictionary contains the following keys:

  • rois: Bounding box coordinates
  • masks: Segmentation masks
  • class_ids: Class IDs for each detection
  • scores: Confidence scores

Post-processing functions include:

  • Drawing bounding boxes on the image
  • Calculating mask pixel sums
  • Segmenting the image based on the highest scoring mask

References