Object Detection Based on the GrabCut Method for Automatic Mask Generation.

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Mask R-CNN object detection is computationally intensive due to manual mask creation during training.
• Automating mask generation is crucial for improving the efficiency of object detection models.

Purpose of the Study:

• To develop an automated mask generation method for object detection.
• To enhance the efficiency of the Mask R-CNN training process.

Main Methods:

• A two-stage approach combining GrabCut for interactive image segmentation and Mask R-CNN for object detection.
• Stage 1: GrabCut generates initial object masks.
• Stage 2: Mask R-CNN utilizes generated masks, original images, and labels for training.

Main Results:

• Achieved a mean average precision (mAP) of over 95% for segmentation on the Berkeley Instance Recognition Dataset.
• Demonstrated a simple and highly efficient method for obtaining segmented target object masks.

Conclusions:

• The proposed GrabCut-based automated mask generation method significantly reduces the time and effort required for training object detection models.
• This approach offers a practical solution for efficient mask acquisition in object detection tasks.