Assignment
For each assignment, you need to fill in all the missing lines of codes, and run *.ipynb to get expected running results. Keep the running results in *.ipynb and submit the folder to canvas.
Deadline: See course schedule for deadline of each assignment. You will be allowed a total of five late days per course. Each additional late day will incur a 10% penalty.
Ethics: Please do assignments individually, while you can search the Internet resources to help you write, debug, and run the codes.
Points: You will get 12/12 points if your submission shows expected running results. Otherwise, you will get fewer points depending on your written codes and running results.
Final Project
In a group of 1 - 2 people (2 is preferred)
Submit a link to project blog post to canvas.
Give a proposal presentation and a final presentation. Submit your slides to canvas.
Topics can be either one of the following. Topics related to deep learning are recommended but not necessary.
Proposing and addressing a new computer vision task/application;
Design and evaluation of a novel approach for a computer vision problem;
An extension of a computer vision approach studied in class;
In-depth analysis of an existing computer vision technique;
In-depth comparison of two related existing computer vision techniques.
For graduate students, moderate methology novelties are expected. For undergraduate students, methology novelties are not required.
Proposal presentation (5 points): A 5-minute presentation describing: (1) Problem statement: describe the problem; (2) Related work: briefly describe related papers, and, what will be unique about your project that previous work has not done? (3) Approach: briefly describe the algorithm(s) you will employ; (4) Experiments: Describe the designed experiments to evaluate your approach. (5) Others: Summarize any preliminary results.
Final project presentation (10 points): A 10-minute presentation describing: (1) Introduction: summarize the problem, main idea, and results; (2) Related work: provide a detailed description of related papers. If you're proposing a new idea or extending an existing approach, what will be unique about your project that previous work has not done? If you're analyzing one or two related techniques, describe how they relate to other relevant work; (3) Approach: Describe in detail the algorithm(s) you employed. Clearly state the method's input and output, and any assumptions or design choices; (4) Experiments: describe the experiments you conducted to evaluate the approach. For each experiment, describe what you did, what was the main purpose of the experiment, and what you learned from the results. Provide figures, tables, and qualitative examples, as appropriate. (5) Conclusions: briefly summarize the main idea and results, discuss limitations and possible future work.
Final project blog post (20 points): The final report is in the form of a blog post. It should be written in a way that is accessible to a general audience, introductory and educative, and include all contents as above. To encourage quality and creativity, the top 20% blog posts as determined by student mutual evaluation (submitting a list of your favorite blog posts to Canvas!) will receive an additional 5 bonus points. Example blogs:
Generative Modeling by Estimating Gradients of the Data Distribution, by Yang Song
Perspectives on diffusion, by Sander Dieleman
Diffusion Models for Video Generation, by Lilian Weng
The AI Revolution: The Road to Superintelligence, by Maarten Steinbuch
Understanding LSTM Networks, by Christopher Olah
The Unreasonable Effectiveness of Recurrent Neural Networks, by Andrej Karpathy