There will be two tracks for course projects: Research and Applications (see details below). Projects can be completed individually or in groups of 2-4. For GPU access, please create an account on Longleaf.
Research Track: The research track is aimed toward (graduate) students pursuing research related to vision transformers. For this track, students can propose any project that involves vision transformers in any area of interest. Undergraduate students interested in vision transformers research who have concrete ideas of what they want to pursue are also welcome to pursue this track but should talk to me before doing so.
Applications Track: The applications track is aimed at (undergraduate) students who want to gain more experience with existing vision transformer tools. Projects in this track will require students to apply existing vision transformer models to 5 of their selected applications (e.g., image classification, action recognition, object detection, object segmentation, image/video generation, etc.). Specifically, for each of the 5 applications, the students will select several images/videos (e.g., personal or from the Internet) and apply a transformer model of their choice to investigate its effectiveness at that particular task. For each application, the students will be asked to provide 3 success instances (i.e., the model did what it was expected to do) and 3 failure instances (i.e., the model failed). This will amount to 15 success and 15 failure instances across 5 applications. Afterward, the students will document their experiences and provide a detailed analysis and insights about the results. Graduate students who think that the undergraduate track fits their background and interests better are welcome to pursue this track. Below are a few potentially relevant pointers to the existing video analysis tools/demos:
You can use the following templates (research and applications) to generate your proposal reports. Your initial proposal should cover the items below and should be ~1 page long.
Research Track:
The problem that you are trying to solve.
The motivation behind the problem.
Your proposed approach and how it relates to prior work.
The experiments that you plan to conduct.
The datasets that you plan to use.
Applications Track:
The 5 applications that you selected.
The 5 video tools you will experiment with and a brief technical description of each tool.
The data that you plan to use.
What you expect to discover.
You should extend your project proposal to ~2-4 pages and include the following:
Research Track:
A preliminary set of results.
Your analysis of those results.
Additional experiments that you plan to run.
Applications Track:
A preliminary set of results (i.e., success and instance cases).
Your analysis of those results.
You should extend your project milestone to ~4-8 pages and include the following:
Research Track:
The final set of results.
Your analysis of those results.
Your overall conclusions and findings from the project.
Applications Track:
The final set of results (15 success and failure instances for 5 applications).
Your analysis of those results.
Your overall conclusions and findings from the project.
We will hold a poster session where each group will present the final results of their projects.
Date: Monday, 05/05/25
Time: 12pm-3pm
Location: Upper Lobby Sitterson
Poster Printing Details: Please upload your poster to Canvas by 10:00 AM on Friday, May 2 and we will print it for you.
Poster Requirements:
Poster size: 24 x 36 inches, either portrait or landscape orientation.
Your poster should include at least the following sections:
Introduction – Provide a brief overview of your project and its motivation.
Method – Describe your approach in detail. Figures or pseudo codes are highly recommended.
Experiments – Show your results, ablations, and comparisons with baselines.
Conclusion – Summarize your contributions and key findings.
Other Poster Resources:
PosterNerd Templates (https://www.posternerd.com/sciposters-templates?srsltid=AfmBOoomJEES7IaGbJIwqGV_W1zxRNmadM7fvEtWH7aYxPuJUi_TlnhK):
Offers 10+ editable templates designed for 24x36'' posters.
Bolei Zhou’s Awesome Posters (https://github.com/zhoubolei/bolei_awesome_posters/tree/main):
Includes examples of posters from top CV/ML conferences. You can refer to them for ideas on content organization, section structure, and visual design.