vinod-polinati/ImageSegmentation

A computer vision Image Segmentation Project done using Segment Anything by Meta.AI & Yolo V8

Object Detection & Orientation using Segment Anything and YoloV8

Utilizing the power of YOLOv8 and Segment Anything by Meta AI, this Python script offers sophisticated template matching and precise orientation detection capabilities, revolutionizing image analysis with cutting-edge technology."

Prerequisites

• Python 3.x
• Segment Anything

      pip install 'git+https://github.com/facebookresearch/segment-anything.git'
  
      wget https://dl.fbaipublicfiles.com/segment_anything/sam_vit_h_4b8939.pth

• YoloV8

    pip install ultralytics

• OpenCV

    pip install opencv-python

• Numpy

    pip install numpy

• ScikitLearn

    pip install scikit-learn

• Matplotlib

    pip install matplotlib

Usage

1. Clone the repository:

   git clone https://github.com/vinod-polinati/ImageSegmentation.git

2. Navigate to the project directory:

   cd ImageSegmentation

3. Run the YoloV8_Sam.ipynb script with the paths to your template and test images

Description

The YoloV8_Sam.ipynb script performs the following steps:

1. Image Loading: Load the template image and the test image.
2. Feature Detection and Description:
   • Use the ORB (Oriented FAST and Rotated BRIEF) detector to find keypoints and compute descriptors for both images.
3. Feature Matching:
   • Match the descriptors of the template image with those of the test image using a Brute-Force matcher.
4. Homography Estimation:
   • Calculate the homography matrix using RANSAC (Random Sample Consensus) to find a perspective transformation between the keypoints.
5. Transform Template Corners:
   • Use the homography matrix to transform the corners of the template image onto the test image.
6. Rotation Angle Calculation:
   • Determine the rotation angle based on the homography matrix.
7. Annotate Test Image:
   • Draw a polygon around the detected object (template) on the test image and annotate the rotation angle.
8. Display Result:
   • Show the annotated test image with the detected object and rotation angle.

EXAMPLE IMPLEMENTATION :

Image used for masking

Detected Image

Masked out Image

Image used to check Orientation

Orientation Degree check

Note

• Ensure that the specified paths to the template and test images are correct.
• The script will display the annotated test image in a window. Press 'Q' to close the window and end the script.

References

• Segment Anything
• YoloV8
• OpenCV Documentation
• Homography Estimation

Feel free to modify and integrate this script into your projects for template matching and orientation detection tasks.