EECS 556 proposal guidelines

Proposal format

Here is some student feedback from a previous term: "The only problem for us is that all of our reports are on the Google Docs and this is not easy to organize the sequences of images, equations and tables, especially when we need to insert a image or equation then the following sequence numbers will all be changed. So next time if necessary we prefer to use online Latex editors."

I will provide an overleaf template for the project proposals so that you can use it to learn about LaTeX and see how easy it is.

Proposal content

Your project proposal should include the following, in this order. This outline is for the recommended Option A.

• 0. Names of all team members, date, course, and concise project title.

  A version of the honor code like this Our team will give attribution for any figures used in our documents and will cite all code sources (beyond those already built into Julia Base or Matlab toolboxes)

• 1. Introduction
  A brief introduction that (i) overviews the image processing problem area, and (ii) discusses how the publication whose results you plan to replicate solves that problem and what are its (supposed) advantages.
  • Some aspect of your project must be based on peer-reviewed work, not just unpublished arxiv paper(s). Journal papers are strongly recommended over 4-page IEEE conference papers. Discuss with me before choosing a conference paper as your focus.
  • Give full bibliographic reference to the paper, including a link (url) to the reference(s).

• 2. Quantitative performance prediction.
  This is a key part of the proposal. It should satisfy the following criteria.
  • The quantitative performance prediction must be crisply defined, and posed in a testable manner.
  • It must correspond to a claimed result in the published study.
  • There must be a reasonable prospect that an investigation this term will be able to assess the performance prediction quantitatively (by replicating evidence in the study).
  • For example, a performance prediction for a paper about an image denoising method might be phrased:

    "The BM3D imaging denoising method yields denoised images that have an average of 3dB higher SNR than methods ABC and XYZ on face images from (some collection) with added noise having standard deviation from 3-7 gray levels."

    This would not be precise enough, though, since it does not specify how many face images will be used from the collection and how the parameters of the BM3D and ABC and XYZ methods will be selected and possibly other assumptions that may be used to assess the prediction. Including these would make this a precise prediction.

  • A prediction like: "We can implement the ABC method of Einstein so that it works well by the end of the term" is not a very scientifically interesting prediction and not quantitative.
  • Try to make your prediction as precise as possible. Think about how it could be evaluated at the end of the term. Ideally, an objective third party should be able to assess the prediction from the evidence you provide.
  • Include any explanatory background needed to understand your prediction. This should not require more than a paragraph or two.

• 3. Plan.
  Describe how you plan to replicate the part of the paper that bears on your performance prediction. Be as specific as possible, recognizing that the exact steps you take will depend on results as you go along. Make note of any requirements, for example obtaining further information, code, or data from the study authors or other source. For team projects, indicate who plans to do what. Explain how your plan will produce the evidence needed to assess the prediction. The plan description should take several paragraphs. Some detail is needed for this part to be complete.
  Give the URL(s) for the code and data you expect to use.

• 4. Extensions (required for Option A, optional for Option B)
  When investigating published methods, student teams usually think of extensions or refinements of the methods that may further improve them in some way (better results, faster computation, simpler implementation, etc.), and this is expected for Option A and should be described here. For Option B, if you already envision one or more possible extensions or alternatives that might interest you if time permits, describe them here. This is your chance to show creativity.

  Alternatively (or in addition), there may be follow-on publications that further improve on the methods in the study you replicate and that you might want to include for comparison if time permits. If so, briefly summarize the method(s) and cite the corresponding publication(s).

  Such extensions are required for (the recommended) Option A, so are a key part of the plan.

• 4. Bibliography.
  See technical writing guidelines.

• Submission.
  One team member uploads to Canvas a file named like this: unique1-unique2-unique3-unique4-proposal.pdf with the last names in alphabetical order.