Skip to content

CS492(A): Machine Learning for 3D Data

Minhyuk Sung, KAIST, Spring 2022

Paper Presentation

Presenter To-Dos

Slide Preparation

Share the link of your slides on Discord by four days before your presentation. You are allowed to edit the slides after sharing the link, but the slides need to be ready to be presented by then.

  • Monday presentation → Share your slides by Thursday 23:59 KST in the previous week.
  • Wednesday presentation → Share your slides by Saturday 23:59 KST in the previous week.

Presentation

Prepare a 20-minute-long presentation. It's strongly recommended to practice your presentation and check whether you can finish in 20 minutes. You may have a 3-minute buffer after spending 20 minutes, and after spending the buffer, your presentation will be cut off.

All team members need to participate in the presentation and speak for roughly the same amount of time. Any team member who does not speak in the presentation will get a zero score. Otherwise, all team members will get the same evaluation score.

Check out the presentation guidelines section below for more details.

[IMPORTANT] Plagiarism

You are allowed to use slides or images from the authors of the paper or any other places on the Internet, but DO NOT FORGET to add references in the slides where you use resources from the other places. The audience should be able to clearly see what are the slides and images that you borrowed from the other places. Do not add the references at the end of the slides but in the place where you use the existing resources.

If references are missing, it will be considered plagiarism, and you will get a penalty in the evaluation.

Back to top


Audience To-Dos

Questions

During each presentation, all audience must post at least two questions on Discord by the end of each presentation. Post your questions as one message on the Discord channel. You'll lose participation scores when you do not post at least two questions.

Group Discussions

Right after each presentation, there will be a 7-8-minute-long group discussion, followed by a 5-minute Q&A session. Each group in a Zoom breakout room will have 6 people including the moderator. The moderators will be randomly selected by TA. The moderator is responsible to lead the discussion and complete the following tasks:

  • Discuss the questions posted in Discord, choose the two most important questions (or think about new questions), and post the questions under the "group-{id}" thread in the Discord channel.
  • Evaluate the presentation and report the score in the group DM sent by the TA.
  • In the group DM, report people in the group who did not participate in the discussion.

For the peer evaluation, check out the evaluation guidelines section below.

Back to top


Presentation Guidelines

The following guidelines are recommended but not required. Think about the most effective way to convey the key ideas in the paper.

  • Describe the motivation of the work based on

    • Downstream applications (e.g., why segmentation is important in autonomous driving?)
    • Limitations of previous work (e.g., what are the limitations of NeRF that have not been handled in previous work?)

  • Think about the best way to hook the audience. Show what has not been done and what becomes available with the proposed technique. Consider showing impressive results of the paper first and explain the algorithm later.

  • Briefly compare the proposed work with previous work and concisely state the difference and contribution of the proposed work (in one sentence).

Algorithm (~10 mins)

  • Describe the exact problem definition. What are the inputs and outputs? What is the given information (supervision) at the training time (for learning-based methods)? What are the assumptions used?

  • If the audience would need background knowledge to understand the proposed method, introduce the background first. E.g., if you introduce PointNet++, explain PointNet first.

  • Describe the algorithm in a top-down way. Provide the overview first and go down to each component in the algorithm.

  • DO NOT explain the algorithm like describing a recipe. Find the key ideas of the paper and focus on explaining them. Describe how the key ideas differentiate the paper from previous work. 20 minutes are not sufficient time to explain all the details. Do not waste your time explaining some backgrounds that everyone would know or some implementation details.

  • Note that you still need to be prepared for all the questions about the details. Consider preparing backup slides.

  • Link your explanation of the algorithm with the later experimental results. Ablation study in the experimental results describes what happens when each component in the algorithm does not exist. Connect your explanation about why it is needed in the algorithm part with your description about what happens if it does not exist in the results.

  • Take full advantage of visual aids. Use more images/videos but fewer texts. Avoid adding more than two sentences per slide. DO NOT just read texts in the slide.

Experimental Results (~4 mins)

  • First, briefly describe the datasets/benchmarks used in the experiments and the evaluation metrics.

  • Briefly describe the baseline methods (the methods that are compared with the proposed method) and show quantitative (tables or graphs) and qualitative (images) comparisons. Highlight the difference with the baselines and describe why the proposed method is better.

  • Describe the ablation study results and connect them with the explanation of the algorithm.

Conclusion / Limitations / Future Works (~2 mins)

  • Summarize the key ideas in ~2-3 sentences.

  • Provide your own critiques. Discuss the limitations of the proposed work and potential research questions to be investigated in future works.

Back to top


Evaluation Guidelines

Evaluate presentations based on:

  • how clearly the presenters delivered the main ideas of the paper;
  • how well the presentation was organized and prepared;
  • how knowledgeable the presenters are about the paper and any related materials; and
  • how much the presenters maintained your interest during the entire presentation.

DO NOT evaluate based on:

  • speech skills (it’s not a speech course),
  • technical issues of Zoom, and
  • whether the presenters used the author’s or the other’s slides (if references exist).

For each group, provide a score on the [1, 5] range (a non-integer number is ok) with any short comments:

  • 5: Probably the best presentation!
  • 4: It was a clear and well-organized presentation.
  • 3: Good, but I could not follow some parts.
  • 2: It could be better.
  • 1: Poor.

DO NOT write the evaluation score in any public channel on Discord. TA will send a group DM to the moderator.

Back to top


Penalty

Presenters

  • Late sharing of slides: 10% penalty in the project evaluation.
  • Sharing incomplete slides: 10% penalty in the project evaluation.
  • Missing references: 10% penalty for each in the project evaluation.

Audience

  • Missing posting at least two questions: Counted as missing a quiz in the participation score.
  • Not participating in a group discussion: Counted as missing a quiz in the participation score.
  • Not playing the role of the moderator: 5% penalty in the paper presentation evaluation.

Back to top