ECE253 Homework 4 Solved

Problem 1. Detecting Objects with Template Matching

Cross-Correlation Filter Read in image birds1.jpeg and template template.jpeg and convert it to grayscale. Perform cross-correlation on the image with the template using convolution and display the resulting image along with a colorbar. You may use library convolution functions.

Normalized Cross-Correlation Apply normalized cross-correlation on birds1.jpeg using tem- plate.jpeg and display the resulting image with a colorbar. Also, display the original image with a rectangular box (the same size as the template) at the location with the highest normalized cross-correlation score. Next, apply normalized cross-correlation using template.jpeg on birds2.jpeg and display the resulting image with a colorbar. Like before, display the original image with a rectangular box at the location with the highest normalized cross-correlation score. Does the box surround any of the birds?

2

Problem 2. Hough Transform

1. (i)  Implement the Hough Transform (HT) using the (ρ, θ) parameterization as described in GW Third Edition p. 733-738 (see ‘HoughTransform.pdf’ provided in the data folder). Use accumulator cells with a resolution of 1 degree in θ and 1 pixel in ρ.
2. (ii)  Produce a simple 11 × 11 test image made up of zeros with 5 ones in it, arranged like the 5 points in GW Third Edition Figure 10.33(a). Compute and display its HT; the result should look like GW Third Edition Figure 10.33(b). Threshold the HT by looking for any (ρ, θ) cells that contains more than 2 votes then plot the corresponding lines in (x,y)-space on top of the original image.
3. (iii)  Load in the image ‘lane.png’. Compute and display its edges with an appropriate threshold.

   Now compute and display the HT of the binary edge image E. As before, threshold the HT and plot the corresponding lines atop the original image; this time, use a threshold of 75% maximum accumulator count over the entire HT.

4. (iv)  We would like to only show line detections in the driver’s lane and ignore any other line detections such as the lines resulting from the neighboring lane closest to the bus, light pole, and sidewalks. Using the thresholded HT from the ‘lanes.png’ image in the previous part, show only the lines corresponding to the line detections from the driver’s lane by thresholding the HT again using a specified range of θ this time. What are the approximate θ values for the two lines in the driver’s lane?

Things to include in your report:

• HT images should have colorbars next to them
• Line overlays should be clearly visible (adjust line width if needed)
• HT image axes should be properly labeled with name and values (see Figure 10.33(b) for example)
• 3 images from 2(ii): original image, HT, original image with lines
• 4 images from 2(iii): original image, binary edge image, HT, original image with lines
• 1 image from 2(iv): original image with lines
• θ values from 2(iv)
• Code for 2(i), 2(ii), 2(iii), 2(iv)

3

Problem 3. K-Means Segmentation
In this problem, we shall implement a K-Means based segmentation algorithm from scratch. To

do this, you are required to implement the following three functions:

• features = createDataset(im) : This function takes in an RGB image as input, and returns a dataset of features which are to be clustered. The output features is an N × M matrix where N is the number of pixels in the image im, and M = 3 (to store the RGB value of each pixel). You may not use a loop for this part.
• [idx, centers] = kMeansCluster(features, centers) : This function is intended to perform K-Means based clustering on the dataset features (of size N × M). Each row in features represents a data point, and each column represents a feature. centers is a k × M matrix, where each row is the initial value of a cluster center. The output idx is an N × 1 vector that stores the final cluster membership (∈ 1, 2, · · · , k) of each data point. The output centers are the final cluster centers after K-Means. Note that you may need to set a maximum iteration count to exit K-Means in case the algorithm fails to converge. You may use loops in this function.
• im seg = mapValues(im, idx) : This function takes in the cluster membership vector idx (N × 1), and returns the segmented image im seg as the output. Each pixel in the segmented image must have the RGB value of the cluster center to which it belongs. You may use loops for this part.

  With the above functions set up, perform image segmentation on the image white-tower.png, with the number of clusters, nclusters = 7. To maintain uniformity in the output image, please initialize clusters centers for K-Means at random.

  Things to include in your report:

  • The input image, and the image after segmentation.
  • The final cluster centers that you obtain after K-Means. • All your code for this problem.

4

Problem 4. Semantic Segmentation (20 points)

In this problem, we will train a fully convolutional network [1] to do semantic segmentation. Most of the code is provided, but after a long day of Digital Image Processing, someone forgot to hit ’save’, so part of the network is missing! Your task is to complete and train the network using the CityScape [2] dataset, and to answer the following questions. Please check the README.md for training and testing commands. (And please, help each other out on Piazza if you get stuck!)

1. Please complete the FCN network, the fcn8s in ptsemseg/models/fcn.py. Briefly describe the model structure.
2. Do we use weights from a pre-trained model, or do we train the model from scratch?
3. Please train the network with CityScape dataset. Visualize the training curves (suggested option: use Tensorboard). Include pictures of the training and validation curve. (config file: configs/fcn8s cityscapes.yml)
4. What are the metrics used by the original paper? Do inference (validate.py) on the valida- tion set. Which classes work well? Which classes do not?
5. Can you visualize your results, by plotting out the labels and predictions of the images? Please
   include at least two examples (HINT: check the unit test in ptsemseg/loader/cityscapes loader.py)
6. Take a photo of a nearby city street, and show the output image from the model. Does the output image look reasonable?
7. Based on your analysis of the model and training, how can you get better results for pre- diction? Give 2 possible options. You may want to think about hyperparameters, network architecture, data, or other considerations. You may also get ideas from the FCN paper [1] and derivative works.

To be noted:

• Upload the zip file to the server. Follow the steps in README.md to install environments

  and requirements

• Training time is around 5 hours.
• When you’re running the server, save the URL so that you can access the tab later once you close it.
• Please read the FCN paper [1].

5

Problem 5. Tritongram (5 points)

With recent news around negative aspects of social media, ECE 253 would like to compete with Meta by making our own, better version of Instagram: Tritongram. While our team of engineers is hard at work building the app, we are on short supply of talented digital image processing experts to create the filters!

This problem is completely open-ended, for a chance to show your creativity and Digital Image Processing skills. The only requirement is that your code (apart from import statements) must be wrapped within a single, well-commented function that takes an RGB image (or list of images) as input, and returns a single RGB image as output.

Please include a demonstration of three sample input/output examples for your filter, and be sure to include a fun title for your filter. Top filters will be added to an ongoing repository on GitHub, which we will share on Piazza at the end of the quarter.

Please do not stress about this problem, we know finals are coming up, this is meant to be enjoyable and easy points. A filter which takes an image and returns an image of 0s will still get full credit, though it would be nice to see something a little more spirited.

References

1. [1]  J. Long, E. Shelhamer, and T. Darrell, “Fully convolutional networks for semantic segmenta- tion,” in 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Jun. 2015, pp. 3431–3440, iSSN: 1063-6919.
2. [2]  M. Cordts, M. Omran, S. Ramos, T. Rehfeld, M. Enzweiler, R. Benenson, U. Franke, S. Roth, and B. Schiele, “The Cityscapes Dataset for Semantic Urban Scene Understanding,” arXiv:1604.01685 [cs], Apr. 2016, arXiv: 1604.01685. [Online]. Available: http: //arxiv.org/abs/1604.01685

6