Deep Learning in Computer Vision

Week 0: Linear Algebra and Calculus

Exercise: Given that x is a vector in R3 and the tail of x is at the origin. Where is the locus of the tip of vector x such that the norm L1 of x is equal to 4. What is the focus if x is a vector in R4?

Week 1: Image Processing in Computer Vision

Exercise:

1. Download any color image file with PNG format from Internet (for those who have no idea about PNG file, please see the link: https://en.wikipedia.org/wiki/Portable_Network_Graphics )

2. Write program to do the following works:

   • Load the color image file downloaded in Step 1
   • Display the color image
   • Convert the color image to a gray image, save to a file
   • Reload the file with gray image and display
   • Make sure your saved files can be opened and displayed by other programs, e.g., ImageViewer, Photoshop etc

3. Take a break

4. Write the program to do the following works:

   • Reload the color image file downloaded in Step 1
   • Resize the image to the size of 256 (pixels) x 256 (pixels)
   • Display the image
   • Save to a file
   • Reload the gray image file converted in Step 2
   • Resize the gray image to the size of 256 (pixels) x 256 (pixels)
   • Display the image
   • Save to a file

5. Write the program to do the following works:

   • Apply Gaussian filter with different kernel sizes and sigma
   • Explain the differences

6. Use the perspective projection equations to explain why, in a picture of a face taken frontally and from a very small distance, the nose appears much larger than the rest of the face. Can this effect be reduced by acting on the focal length?

Week 2: Traditional Machine Learning

Exercise:

1. Download the Iris flower data set (https://en.wikipedia.org/wiki/Iris_flower_data_set)

2. Write program to complete the following works:

   • Visualize the dataset
   • Build a decision tree classifier to classify this dataset
   • Modify parameters/hyper-parameters to get the best result

3. Take a break

4. Write the program to do the following works:

   • Build a SVM classifier to classify this dataset
   • Modify parameters/hyper-parameters to get the best result

5. What’s the trade-off between bias and variance?

6. What is the difference between supervised and unsupervised machine learning?

7. How is KNN different from k-means clustering?

Week 3: Edge Detection, Image Tracking

Exercise:

1. Download any color image from Internet and save it to your computer

2. Write program to complete the following works:

   • Convert the downloaded image from 1. to a grayscale image
   • Apply Canny Edge detector to the grayscale image with fixed threshold as your choice (https://en.wikipedia.org/wiki/Canny_edge_detector)
   • Write a small application to find the Canny edge detection whose threshold values can be varied using two trackbars

3. Take a break

4. Review the watershed algorithm (https://en.wikipedia.org/wiki/Watershed_(image_processing))

5. Write program to complete the following works:

   • Download any color image from Internet and save it to your computer
   • Convert the downloaded image to a grayscale image
   • Apply watershed algorithm to the above image and observe outputs with different parameters

You can refer to the following to understand more about this algorithm: http://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_imgproc/py_watershed/py_watershed.html.

Week 4: Tensorflow/Keras

Exercise:

1. Install Tensorflow. Instruction: https://www.tensorflow.org/install/
2. Perform image recognition using pre-trained models. Instruction: https://www.tensorflow.org/tutorials/image_recognition Try different networks:
   • AlexNet
   • VGG
   • Inception
   • ResNet

Week 5: Tensorflow exercise

Exercise: Using Google Colab, open the Jupiter notebook file, try to modify each layer's hyperparameters, dropout connection between layers to get better model.

Week 6: Object detection / Sematic segmentation

Week 7: Transfer learning / Fine-tuning

Week 8: Applications

Week 9: Transfer learning tutorial

Lecture URL : https://www.youtube.com/watch?v=2XKh4BtzUMo