CAP5415-Computer Vision (FALL 2018)

Instructor: Dr. Sedat OZER    

• Neural Networks
• Convolutional Neural Networks
• RNN, LSTM-based Deep Networks for sequences in vision
• Deep Reinforcement Learning
• Optimization and hyperparameter tuning for vision
• Image Filtering, Edge Detection
• Object Classification (LeNet5, AlexNet, VGG16, ResNet, InceptionNet)
• Object Detection (R-CNN, Fast R-CNN, Faster R-CNN, YOLO)
• Segmentation, Semantic Segmentation
• Hand-crafted Features (SIFT)
• Interest Point Detection
• Optical Flow

• Final project: 25% (+5% for the progress report. The progress report format will be available on the webcourses). Projects will be presented at the end of semester.
• Mid-Term Exam 25%. (tentative date: First week of October 2018, in-class, written)
• Paper presentation: 20% of the final grade.
• See the guidelines on webcourses for both the final project and paper presentations.

• Ian Goodfellow, Yoshua Bengio and Aaron Courville, Deep Learning, MIT Press, 2016.
• Simon Prince, Computer Vision: Models, Learning, and Interface, Cambridge University Press,
• Mubarak Shah, Fundamentals of Computer Vision,
• Richard Szeliski, Computer Vision: Algorithms and Applications, Springer, 2010 (online draft),
• Forsyth and Ponce, Computer Vision: A Modern Approach, Prentice Hall, 2002,
• Palmer, Vision Science, MIT Press, 1999,
• Duda, Hart and Stork, Pattern Classification (2nd Edition), Wiley, 2000,
• Koller and Friedman, Probabilistic Graphical Models: Principles and Techniques, MIT Press, 2009,
• Strang, Gilbert. Linear Algebra and Its Applications 2/e, Academic Press, 1980.

LECTURE NOTES

• Tutorial: Introduction to Eigenvalues and Eigenvectors (by G. Strang)
• Week 1: Introduction to Computer Vision
• Week 2: Introduction to Classification, Logistic Regression and Gradient Descent algorithm
• Week 3: Introduction to Fully Connected, Multi-layer (Feed-forward) Neural Networks, SoftMax and Activation Functions
• Week 4: Introduction to Convolutional Neural Networks, Convolution operation, Edge detection and Filtering
• Week 5: Continue on ConvNets: Image Classification (LeNet5, AlexNet, VGG-16, ResNet, InceptionNet)
• Week 6: Object Localization and Detection with ConvNets: Region Proposals and R-CNN, Fast and Faster R-CNN
• Week 7: Recitation and Midterm
• Week 8: Continue on Object Detection with ConvNets: Sliding windows and YOLO algorithm
• Week 9: Optimization and hyperparameter tuning
• Week 10: Group Paper Presentations
• Week 11: Classification with hand-crafted features: Part 1: Corner (curvature) detection, Harris features, SIFT algorithm, Keypoint matching , Transformations - Homography and Image stitching
• Week 12: Classification with hand-crafted features: Part 2: Bag-of-words representations with K-Means, Classification with kernel machines: Support Vector Machines (SVM), Multiple Kernel Learning (MKL) and Similarity Domains Machine (SDM)
• Week 13: Segmentation vs. Semantic Segmentation, Optical Flow
• Week 14: Introduction to RNNs, LSTM for sequence learning in vision and Deep Reinforcement Learning for Visual Tracking

PAPER PRESENTATION

Please submit your paper presentation details as text first. Each project group will present one paper.

In your text submission, you need to include:

Paper's info:
• Full title of the paper as shown on the paper,
• Complete author list of the paper,
and where the paper is published (peer-reviewed) along with its publication year,
• Paper's abstract,
• Paper's link to download,
• Also include your own project title and your project's one sentence description in another paragraph.


The paper presentation will take 20% of your final grade. Please choose a paper relevant to your project title. That 20% distributes as follows:


• Paper proposal submission: 3% ,
• Presentation: 17%,
• Presenting "the content" properly: 12%,
• Presentation quality: 5%,

Presentation details: For "the content" of the paper that you picked, you will focus on the following details and will present them in your presentation:


• Check for the code of the paper on GitHub or on another online platform and download and run the code to generate author's results. (4%)
• Understand the goal of the paper: What is new, and what was done previously as described in the paper. (2%)
• What is the overall solution and the architecture presented in the paper, (2%)
• What metrics used in the paper and their definitions, (2%)
• What are the results, and how they are presented. (2%)
• BONUS: apply the code on another data-set that was not used in the paper and compute the performance metrics used in the paper. (+%3 on your final grade)
• Expect to have a short presentation focusing on the above-listed points.

Final-Project Topics

• Autonomous Path Planning with Deep Reinforcement Learning
• Face Detection for Android Applications with Deep learning
• Image inpainting with GANs
• Object detection with YOLO and Faster-RCNN
• Object/Actor segmentation with Deep Learning
• Object Tracking in Videos with Deep Learning
• Video generations with GANs
• Scene understanding /parsing with GANs
• Person re-identification with GANs
• Activity Recognition/Detection in videos with Deep Learning

Contact