3. Submission Guidelines

https://openreview.net/group?id=kaist.ac.kr/KAIST/Spring2021/CS492H

If you don’t have an account at OpenReview, sign up here:
https://openreview.net/signup
Please use your KAIST email address.

• A submission must include a report, a poster, and a code repository link. (Supplementary materials are optional.)
• Submission with the wrong format or any missing items may receive a zero score.
• Make a submission for each team (not for each person). Add all the team members’ names in the submission.
• Development Track: Make the title to be the same as the original paper.
• In the OpenReview webpage, the deadline ie written as Jun 06 2021 02:59PM UTC-0), which is the same as Jun 06 (Sun), 23:59 KST.
• Even in the case when OpenReview does not block the submission after the deadline (due to some system errors), any submissions after the deadline will be rejected.

Report Guidelines

Format

• A single PDF file (maximum 20MB). Supplementary materials are allowed.
• Must be in CVPR format:
  LaTeX/Word Templates Zip file (CVPR 2021 format)
  Overleaf template (CVPR 2018 format)
• Add the project type (Development Track / Research Track) in the second line of the report title. Using LaTex, you can add like this: \title{Project Title \\ {Development/Research} Track}.
• Must be up to four pages long, including figures, tables, and acknowledgments, but not references. Additional pages containing only cited references are allowed.

Please use the same section names below to facilitate the review process.
Additional sections/subsections are allowed. Also, the detailed instructions in each section are recommendations, and you do NOT need to exactly follow them, except for the highlighted parts.
The review will be single-blind. Add author names in the report (uncomment \cvprfinalcopy in the CVPR template).

• DO NOT COPY the texts or figures in the original paper (except for qualitative result figures) — it will be considered plagiarism.

  Write the report in your own words. It is allowed to use qualitative result figures in the original paper for comparisons (actually highly recommended), but clearly mention that the figures are from the original paper.

• Make the title to be the same as the original paper.

  
  

• Abstract
  • one-sentence TL;DR,
  • 2-3 sentences for the motivation and key ideas of the original work (add a citation),
  • a few sentences summary of the implementation challenges and your approaches, and
  • 1-2 sentences summary of the experimental results.
  • In the end, add “Code is available at: {link}.”
• Introduction
  • What is the motivation for the (original) work? Why is it important to solve the problem? What would be the potential impacts of this work in the research field, or what would be the potential applications?
  • What is the problem setup the original work introduces? (Add a citation of the original work.) What are the input and desired output? What are the given information in the training time and the given assumptions? What are the main ideas of the original work to solve the problem?
  • What are the key challenges in implementing the proposed method? Does the framework include some components that are not implemented in public libraries (e.g., PyTorch, PyTorch3D)? Do you implement some components from scratch without the public libraries? (Using CUDA?) Is the framework complicated with lots of components? Are there technical details hidden in the paper/code? Doesn’t the authors provide their code or datasets? Does the authors’ code even not reproduce the results in the paper? Is the network training tricky or unstable? Or, do you improve the original work?
  • Summarize the implementation details. Clearly differentiate the parts you implemented and the parts you borrowed existing code or got some help from the external collaborators. How do you implement each part, and how do you handle the challenges? A teaser figure that illustrates the parts you focused on in the implementation would be great.
  • Briefly describe the experimental setups. Which datasets/benchmarks do you use? How do you evaluate the results? Do you make any changes in the experimental setups of the original work? If yes, why? How good are your results compared with the results reported in the original paper (or the results of the authors’ code)? Do you improve the original method, or do you conduct additional experiments not shown in the original work?
  • A summary of achievements at the end is also recommended (in a bullet point list). You can start the summary with (for example) ‘In summary, our achievements are:’.
• Method Summary
  • Describe the problem setup again, but in detail and also in a formal way. Introduce essential math notations for the following formulations. DO NOT COPY the texts or figures in the original paper, but use the same math notations in the original work.
  • Briefly summarize the method proposed in the original work, while focusing on the parts that you implemented (if you used existing codes for some parts). The report needs to be self-contained; the readers must be able to understand the main ideas of the original work without reading the paper.
• Implementation Details
  • First, briefly but clearly state which parts you implemented in the framework. If you used some existing codes, describe in detail how you also adopted the existing codes. Add references for the existing codes. Hiding the use of existing codes will be considered plagiarism.
  • Data generation/preprocessing and evaluations are also parts of the implementation. If you implemented these yourself, illustrate the details.
  • Consider as you explain the implementation details to help readers reproduce the results (like providing a recipe in cooking). Make the exposition easy to follow, but also clear and specific as much as possible.
  • Use visual aids in your exposition. More would be better. But DO NOT COPY figures in the original paper. Formulations are fine (but do not copy and paste an image but write them yourself).
• Experimental Results
  • Clearly describe the experiment setups, including benchmarks/datasets and evaluation metrics. If you change or simplify the experimental setups of the original work, explain why.
  • Provide comparisons with the original work. Use the exact results reported in the original paper for comparison. If you test the authors’ code in new datasets, explain why. Provide both quantitative results (tables) and qualitative results (figures). Explain in which aspects the results of your method is better or comparable.
  • Try to reproduce all the results in the original paper. Any lack of experiments, evaluations, or comparisons (without any appropriate reasons) will be penalized in the evaluation.
  • Make fair and reasonable (apples-to-apples) comparisons. Unfair benefits to your implementation in the experiments will also be penalized.
  • If you achieve better performance than the original work, if you improve the proposed algorithm, or if you conduct some additional experiments not performed in the original paper, emphasize the results.
  • In here or in the conclusion, consider showing some remarkable failure cases explaining the limitation of the proposed method or your implementation, or motivating some future works.
• Conclusion
  • Briefly summarize the project, particularly the main parts you implemented and the experimental results.
  • Describe the limitation of the proposed work or your implementation and potential future research directions (if you have space in the report).
• Acknowledgments
  • Make acknowledgments as a subsection without a section number or as a paragraph.
  • Describe the role of each team member and the external collaborators and their contributions.
  • Hiding the external collaborators will be considered academic misconduct.
  • Acknowledgments section is required, even when you do not have the other team member, external collaborators, or any data/codes you borrowed from the other places (state the fact).

• Abstract
  
  • one-sentence TL;DR,
  • 1-2 sentences for the motivation of the work,
  • a few sentences summary of the problem statement and key contributions, and
  • 1-2 sentences summary of the experimental results.
  • In the end, add “Code is available at: {link}.”
• Introduction
  • What is the motivation for the work? Why is it important to solve the problem? What would be the potential impacts of this work in the research field, or what would be the potential applications?
  • What are the challenges in the problem? How has previous work approached this problem, and what are the limitations (add citations)? Or, why hasn’t the problem been addressed while it is important?
  • What is the problem statement? What are the input and desired output? What are the given information in the training time and the given assumptions? A teaser figure that effectively illustrates the problem setup or the desired output would be great. (E.g., show the best result with the input.)
  • What are your key ideas for solving the challenges? It would be also great if your teaser image also describes the key ideas. (E.g. how do your key ideas make differences in the results?)
  • Briefly describe the experimental setups. Which datasets/benchmarks do you use? How do you evaluate the results? What is the conclusion? (E.g. our method outperformed SotA methods by huge margins.)
  • A summary of contributions at the end is also recommended (in a bullet point list). You can start the summary with (for example) ‘In summary, our contributions are:’.
• Related Work:
  • Consider making two or three groups of related papers and writing a short paragraph for each of them, which a paragraph title. E.g., if your project is about ‘3D GAN’, you can consider groups of ‘Deep Learning for 3D Data’ and ‘Deep Generative Models’. Also, an intro sentence briefly describing the groups is recommended (but not necessary).
  • Summarize each work with 1-2 sentences while focusing on describing in which aspects your work is different or overcoming the limitations.
• Method
  • Describe the problem setup again, but in detail and also in a formal way. Introduce essential math notations for the following formulations. A figure describing the overall method is recommended.
  • The report needs to be self-contained; the readers must be able to understand the ideas without reading the other papers. If needed, provide background knowledge (the ‘minimal’ background to understand your work).
  • Think about the most effective way to explain the key ideas. One option would be prioritizing components in your method not based on the order in the method but based on the importance. Less important details can go to the end or to the supplementary.
  • Use visual aids in your exposition. More would be better.
  • Make the exposition as clear and specific as possible. Formulations can make the exposition clearer and more concise.
• Experimental Results:
  • Clearly describe the experiment setups, including benchmarks/datasets, evaluation metrics, and baseline methods.
  • Provide both quantitative results (tables) and qualitative results (figures). Provide comparisons with the baseline methods. Explain in which aspects your method is better. Think about the best way to demonstrate the advantages of your method over the baseline methods.
  • Conduct the ablation study and show the results. If you have multiple technical contributions, demonstrate how each component affects the results.
  • Do not miss any well-known benchmarks/datasets, standard evaluation metrics, or previous methods. Any lack of experiments, evaluations, or comparisons will be penalized in the evaluation.
  • Make fair and reasonable (apples-to-apples) comparisons. (Or, if the baseline methods take some advantages such as stronger supervision, describe clearly.) Unfair benefits to the proposed method in the experiments will also be penalized.
  • In here or in the conclusion, consider showing some remarkable failure cases explaining the limitation of the proposed method or motivating some future works.
• Conclusion
  • Briefly summarize the project, particularly the key ideas and the experimental results.
  • Describe the limitation of the proposed work and potential future research directions (if you have space in the report).
• Acknowledgments
  • Make acknowledgments as a subsection without a section number or as a paragraph.
  • Describe the role of each team member and the external collaborators and their contributions.
  • Hiding the external collaborators will be considered academic misconduct.
  • Acknowledgments section is required, even when you do not have the other team member, external collaborators, or any data/codes you borrowed from the other places (state the fact).

Poster Guidelines

Format

• A single PNG file (maximum 3MB).
• Must have 2560x1280 resolution.
• Poster template: Link (Design adapted from Zoya Bylinskii)
  Do not need to use this template.
• Must include project title, project track, team member names, and acknowledgments.
• Must only include the figures and experimental results shown in the report. Any new figures and new experimental results must not be shown on the poster. It is allowed to edit the figures and tables.

Resources

• How to Make a More Effiective Research Poster (Zoya Bylinskii)

Code Submission

• Create a public code repository in GitHub or Bitbucket and submit the link.
• Also, add the link in the abstract of the report (see Report Guidelines).
• All the results shown in the report must be replicable with your code.

  In the code repository, provide links of the pretrained models (stored in Google Drive or any other places) and also datasets you created, so that anyone can easily check the replicability.

Supplementary Materials

• Supplementary materials are optional.
• Any formats (PDF document, video, web page) are allowed (maximum 100MB).
• Submit as a ZIP file.

Plagiarism / Academic Misconduct

Plagiarism consists of appropriating someone else’s ideas or any texts, figures, or results of the other work, without sharing credit or correctly citing or mentioning the work. If plagiarism or other academic misconduct is discovered in your submission (in any of your report, poster, code, or supplementary), you will receive an F grade, and also it will be reported to the university.

Back to top