Final project

The goal of the final project is to get in-depth experience in one specific area in the field of deep-learning. You are free to choose any project as long as it involved a deep network. You may work in a group of up to 4 students.

The project may cover your research, but can not intersect with other class projects.

Timeline

• Project proposal (5%): Due Sun Oct 3.
• Preliminary report (5%): Due Sun Nov 21.
• Final report (25%): Due Sun Dec 5.
• Project presentation (5%): Nov 30 or Dec 02

Project proposal

Your project proposal should a paragraph describing:

• What problem you want to tackle
• What network you want to train
• What data you will use
• Any relevant baselines (prior work)
• A plan to evaluate your model

Each criterion has a weight of 2pt with a bias of -5pt (for a max total of 5pt).

Submission on canvas

Project milestone

Write a report 2 pages in latex cvpr format. Your report should cover

• Introduction: Motivation of your problem
• Problem statement (part of intro section): Motivation and high level overview of your solution
• Related work: Prior work and your relationship to it
• Technical approach: A detailed overview of your technical approach
• Preliminary results: State results you already obtained, and experiments you would like to still run

Each criterion has a weight of 2pt with a bias of -5pt (for a max total of 5pt).

Submission on canvas

Final report

Write a report 4 pages in latex cvpr format. Your report should cover

• Introduction: Motivation of your problem
• Problem statement (part of intro section): Motivation and high level overview of your solution
• Related work: Prior work and your relationship to it
• Technical approach:
  • Originality (with respect to published work)
  • Fit (Is your approach a good solution to the problem)
  • Breath (Did you consider alternatives)
• Results:
  • Discuss your experiments
  • Ablate your method (experimentally verify that your method is better than alterantives)
  • Compare to prior work
• Conclusion

Each criterion (sub-criterion) has a weight of 5pt with a bias of -25pt (for a max total of 25pt).

Submission on canvas

Please indicate external authors or contributors.

Project presentation

5 min presentation

• Introduce your topic, give some context (~1min)
• Talk about your technical approach (1-2min)
• Talk about alternatives (<1min)
• Talk about results (1-2min) 1 min Q/A

Please stick to the time limit. Presentations may be smoother if only one person presents (which is perfectly fine).