MASK R-CNN: A UNIFIED FRAMEWORK FOR OBJECT INSTANCE SEGMENTATION AND BEYOND
This repo tackles object instance segmentation, specifically training Mask R-CNN to detect and segment balloons in the COCO dataset.
Highlights:
- Fine-tuned Mask R-CNN for accurate balloon detection.
- Leverages pre-trained model for efficient learning.
- Optimized hyperparameters for balloon task.
- Evaluated with metrics like AP and MSE.
- Visualizations showcase generalization to unseen images.
Getting Started:
- Clone & install dependencies.
- Train & evaluate model (
train.py
,evaluate.py
). - Visualize results (
visualization.py
).
Dataset & Pre-trained Model:
- COCO dataset for training/validation.
- "COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x.yaml" model architecture.
Configuration & Results:
MODEL.ROI_HEADS.NUM_CLASSES
set to 1 for balloons.- Hyperparameters fine-tuned for improved accuracy (details in
config.yaml
). - AP: [Insert your score]. MSE: [Insert your score].
Visualization:
Balloon detection & segmentation examples in the results
folder.
Contribute & Share:
Fork & improve this project! Share your contributions!
This shorter version keeps the key points while reducing the overall length. Feel free to adjust it further based on your specific needs and preferences.