Introduction to Computer Vision (ECSE 415) Assignment 5: Segmentation
Please submit your assignment solutions electronically via the myCourses assignment dropbox. The submission should include a single Jupyter notebook. More details on the format of
the submission can be found below. Submissions that do not follow the format will be penalized 10%.
The assignment will be graded out of a total of 100 points. There are 50 points for accurate analysis and description, 40 points for bug-free and clean code, and 10 points concerning the appropriate structure in writing your report with citations and references.
Each assignment will be graded according to defined rubrics that will be visible to students. all parts of the assignment except those stated otherwise. Students are expected to write their own code. (Academic integrity guidelines can be found here).
Assignments received late will be penalized by 10% per day.
Submit a single Jupyter notebook consisting of the solution of the entire assignment.
Comment your code appropriately.
Give references for all codes which are not written by you. (Ex. the code is taken from an online source or from tutorials)
Do not forget to run Markdown (’Text’) cells.
Do not submit input/output images. Output images should be displayed in the Jupyter
Make sure that the submitted code is running without error. Add a README file if required.
If external libraries were used in your code please specify their name and version in the README file.
We are expecting you to make a path variable at the beginning of your codebase. This should point to your working local (or google drive) folder.
Ex. If you are reading an image in the following format:
img = cv2.imread ( ’/content/drive/MyDrive/Assignment1/images/shapes.png’ )
Then you should convert it into the following:
path = ’/content/drive/MyDrive/Assignment1/images/’ img = cv2.imread(path + ’shapes.png’)
Your path variable should be defined at the top of your Jupyter notebook. While grading, we are expecting that we just have to change the path variable once and it will allow us to run your solution smoothly. Specify your path variable in the README file.
Answers to reasoning questions should be comprehensive but concise.
1 K-Means and Mean-Shift Clustering for Segmentation (50 points)
In this section, you will be asked to compute image segmentations by using several basic clustering techniques. Clustering is used to determine the class of each pixel, and the result can be different depending on the feature space. The images for this part are placed under the same dictionary.
1. ComputethefeaturesofthePerson.jpgandLandscape.pngimagesbyconvolvingtheimages with the two Haar filter kernels shown below. The white areas of the Haar filter kernel all have a weight of +1, while the black areas have a weight of -1. For the purposes of outside the borders of the image are 0. You could use the integral image technique to implement the Haar filtering in a more computationally efficient (i.e. faster) manner.
Display the filtered feature images.
(a) Rectangle with size 24x12 pixels. (b) Square with size 24x24 pixels. Figure 1: Haar Filters for computing image features.
2. Implement the K-means clustering to compute the segmentation of the Person.jpg and the Landscape.png image with Haar features. Set K=3. Display the segmented images.
3. Implement the Mean-shift clustering to compute the segmentation of the Person.jpg and
Landscape.png images. Display the segmented images.
You can use the scikit implementation of the mean shift method:
(a) Person image for part 1,2. (b) Landscape image for part 3. Figure 2: Images for segmentation.
4. Discuss the benefits and limitations of these clustering methods for image segmentation. 2
Figure 3: Street view for segmentation.
2 Neural Network Implementation for Image Segmentation (50 points)
There are several neural networks widely used for object detection and image segmentation. In this assignment, you will be asked to use a pre-trained with the object category. The network also provides the instance level segmentation of the object inside each bounding box. For more information, please refer to the GitHub repository: Mask R-CNN for Object Detection and Segmentation.
1. Implement the pre-trained Mask R-CNN model and run it on the street.png image included in the assignment folder.
2. Display the result that shows the bounding boxes, object classes, and segmentations inside each bounding box.
3. Repeat steps 1 and 2 for an image of a Montreal street scene that you took with your own camera. You can use the image that you acquired for Assignment 4.
4. Evaluate the performance of this model and explain the steps that this network took to achieve the final result.