29 / 10

participants / team

Topic provider

Advising Organization：Ocean Affairs Council

Organizer

Co-organizer

Co-organizer

2026/10/06 複賽競賽結果公告(更正公告)

【重要更正】最佳輕量化獎第二名是BlueTrace
【Important Correction】 The 2nd place winner of the Best Lightweight Optimization Award is team BlueTrace.

感謝所有參賽隊伍與選手的熱情投入與精彩表現！本屆「2026 海廢影像辨識國際 AI 競賽(複賽)」複賽評選工作已圓滿結束。評審團依據實行可行性、完整度及說明文件 README.md 包含環境建置步驟、執行模型訓練與推論之指令說明等表現指標，經過嚴謹客觀的綜合評審，最終選出晉級決賽隊伍與模型輕量化獎名單。

We would like to extend our sincere gratitude to all participating teams and contestants for their enthusiastic involvement and outstanding performance! The final round evaluation for the 「2026 Marine Debris Image Recognition International AI Competition (Final Round)」 has successfully concluded.

Based on performance metrics including implementation feasibility, overall completeness, and documentation (specifically the README.md covering environment setup steps, as well as instructions for model training and inference execution), the review panel conducted a rigorous and objective comprehensive assessment to determine the winning teams and the Lightweight Model Award recipients.

獲獎團隊(Winning teams)：

RankingTeam NameScore
1st Place金采源是我老婆0.8159878
2nd Place我這一生如履薄冰0.8080843
3rd Placechmod 7770.7991160


模型輕量化獎(Best Lightweight Optimization Award)：

NumberTeam NameScore
1金采源是我老婆-20.16
2BlueTrace-20.21

再次感謝各位選手為海洋環境保護與 AI 應用創新所貢獻的能量與努力！頒獎典禮的時間會另做通知，敬請期待。

Once again, we extend our deepest gratitude to all contestants for their dedication and contributions to marine conservation and AI innovation. The schedule for the awards ceremony will be announced in due course. We look forward to celebrating your achievements.

Introduction

As global attention on marine debris intensifies, leveraging AI to enhance governance efficiency has become a prominent international trend. To accelerate this digital transformation, the National Academy of Marine Research (NAMR) has established an AI-Ready marine debris dataset covering 33+1 distinct categories. To this end, NAMR is hosting the "International AI Challenge on Marine Debris Image Recognition," officially releasing this high-quality dataset this July as a training set for machine learning models. The competition openly invites global talent from fields such as artificial intelligence, computer vision, and marine science to step up to the challenge. By developing high-precision object detection models, participants will turn technological innovation into tangible environmental impact, working together to shape a smarter, more sustainable future for marine governance.

Eligibility

  • Open to high school students, university/college students, and professionals nationwide. Teams are not required to include non-citizen members; however, teams that include non-citizens (excluding citizens of the People's Republic of China (PRC)), and not limited to the Indo-Pacific region) are encouraged to apply.
  • Teams with a higher proportion of non-citizen members will be given priority in selection. Teams must submit resumes, past notable achievements, and background information of members during registration for the preliminary round. The organizer will review these submissions based on the aforementioned priority principle to determine the final list of participants.

How to Register

  • The top 10 teams from the preliminary round will qualify for the finals.
  • Team size and members must remain the same as in the preliminary round.
  • If you have any questions, Please contact the International Marine Debris Image AI Recognition Challenge Organizing Team at mdivrc@digitimes.com

Prize

Total prize pool: NT$300,000

PlaceNumber of WinnersPrize Money
1st Place1120,000
2nd Place1100,000
3rd Place180,000

Best Lightweight Optimization Award (2 winners): Competition Certificate

Activity time

All event times are based on Taiwan Time (GMT+8).

DateEvent
2026/09/03Announcement of Finalists
2026/09/06Final Technical Workshop
2026/09/06 09:00:00Test Set Release and Final Submission
2026/09/20 23:59:59Deadline for Final Submission
2026/10/02Winners Announcement

Supplementary Provisions

  1. The "2026 International Marine Debris Image AI Recognition Challenge" (hereinafter referred to as "the Event") is organized by the National Academy of Marine Research (hereinafter referred to as "the Organizer"). Upon completion of registration, participating teams and their members are deemed to have accepted the competition regulations, rules, announcements, and judging results published by the Organizer. In the event of any violation, the Organizer reserves the right to disqualify the team and revoke its prize eligibility. Those unwilling to accept these terms may freely withdraw from the selection process before the evaluation begins, provided that they notify the Organizer in writing.
  2. Participating teams guarantee that all information provided, filled in, and competition content submitted are:
    • True and free from any infringement of others' rights (including but not limited to portrait rights, copyrights, and/or other intellectual property rights).
    • Original and do not involve plagiarism or ghostwriting. If any falsehood or infringement is discovered or reported and confirmed, the Organizer may disqualify the entry during the event. For winning entries, the award status will be revoked, and the winners must return the prize money and certificates. If a third party claims rights against the Organizer, the participating team shall be responsible for resolving the dispute and bearing all losses and expenses incurred by the Organizer (including but not limited to attorney fees, litigation costs, settlement fees, etc.), which have no relation to the Organizer. In case of disputes over rights and the chronology of creation, the participating team bears the burden of proof and shall raise no objection to the Organizer’s final decision.
  3. Competition entries must not be repetitive works that have previously won awards in domestic or international competitions, and the same work must not participate in other domestic or international competitions during this competition period.
  4. The intellectual property rights of the entries in this competition belong to the participating teams. Participants enjoy the right of first refusal to negotiate cooperation with the proposition enterprise regarding the proposal itself and derived intellectual property rights transfer or licensing matters.
  5. If the maximum number of participating teams is exceeded, admission will be determined based on the registration timestamp.
  6. This competition designates AWS as the development environment, providing an exclusive workspace for the marine debris competition, which will be open during the competition period. After the competition concludes, the organizer/ host will terminate the AWS cloud service environment provided for the competition. The organizer/host is not responsible for backing up or preserving your data on the platform; please back up your files on your own before the event ends.
  7. If a participating team's entry involves violence, pornography, defamation, insults, or other content detrimental to social morality or social justice, the organizer/host reserves the right to terminate the entry's participation.
  8. All teams entering the finals and the allocation of awards will be decided by the judging panel based on the quality of the entries. If necessary, slots may be "added" or "left unawarded", and the panel may also decide to alter award names.
  9. Participating teams agree to license materials provided for the competition (including but not limited to solution presentations, videos, and other outputs) to the Organizer free of charge for various promotional purposes related to this event. If a third party's intellectual property or other rights (including but not limited to portrait rights) are used in the competition content, the participating team must confirm and warrant that there is no unlawful infringement, and that they have obtained the third party's permission to sub-license the materials to the Organizer for an indefinite period within the Taiwan region, utilizing methods including but not limited to public transmission, distribution, public display, and public publication.
  10. Participating teams and their members should act in a manner that maintains the Organizer's reputation. If any issues arise, please contact the project staff proactively to seek a resolution. No defamation should occur before facts are clarified.
  11. The organizer/host will fulfill its duty of confidentiality regarding the personal data filled out by participants and will never leak it. Please fill it out with peace of mind.
  12. Registering for this competition through this event website indicates that you have agreed to the Personal Data Authorization Terms of this event.
  13. Winners must provide complete award-receiving documents required by the Organizer and complete the award process within the designated time. Those who fail to complete the procedures within the specified time will be deemed to have waived their winning qualifications.
  14. According to the tax laws of the Republic of China, if a participating team is subject to a mandatory withholding tax of 10% to 20% on the prize received (the prize money provided by the Organizer is tax-inclusive), and the winner fails to pay the tax in accordance with regulations, the Organizer may cancel their prize eligibility. Furthermore, the Organizer will not interfere with the internal distribution of prizes within the team.
  15. AI Model and Source Code Licensing: The top three winning teams must grant a non-exclusive license for the intellectual property rights of their AI model files and source code to the Academy for academic, public welfare, and official purposes. Under no circumstances will it be used for commercial purposes.
  16. The Organizer reserves the right to modify, suspend, or cancel the content of this event. For matters not covered herein, they shall be handled in accordance with the Organizer's relevant regulations or interpretations, and supplementary announcements may be made at any time.Personal Data Use Policy: Participants agree that DIGITIMES and partner units (including organizers, co-organizers, technical support partners, and partners) will jointly retain your personal information. In accordance with the Personal Data Protection Act,  for the purposes of executing, advertising, and marketing of the "2026 International AI Challenge for Marine Debris Image Recognition", personal data provided by participants will be collected, processed, and utilized within the Taiwan region until the participant proactively requests the organizer to delete or stop processing and utilizing their personal data. Participants have the right to request the organizer to access, provide copies of, or correct their personal data at any time, and may also contact the organizer at any time to object to the continued collection, processing, utilization, or deletion of their personal data. If a participant disagrees to provide or provides incorrect personal data, the organizer will cancel the participant's eligibility to participate or win. For any questions, please contact the "International AI Challenge for Marine Debris Image Recognition" Working Group at mdivrc@digitimes.com.

Evaluation Criteria

The evaluation metric uses mean Average Precision (mAP)[1] at an Intersection over Union (IoU)[2] threshold of 0.5. A prediction bounding box is considered a True Positive (TP) if its IoU with the ground truth bounding box is greater than 0.5; otherwise, it is a False Positive (FP). Precision is then calculated based on TP and FP counts. The system evaluates the AP score for each object type and then averages the AP values across the 19+1 classes of marine debris objects to obtain the final mAP evaluation value, which determines the ranking.

The system uses the COCO API[3] to calculate the mAP values.

Reference
[1] Average Precision (AP): Average precision
[2] intersection over union (IoU): Jaccard index
[3] COCO API: https://github.com/cocodataset/cocoapi

The Best Lightweight Optimization Award uses NetScore as the comprehensive evaluation metric. All teams that advance to the semi-finals and submit a valid model meeting the regulations automatically qualify for this award.

The evaluation formula is as follows:

$$\Omega = 20\log _{10}(\frac{a^2}{(\sqrt{p}\sqrt{m} )} )$$

  • $a$（Model Accuracy）: Measured as a percentage of mAP. For example, if the mAP is 0.85, then $a$=85
  • $p$（Model Parameter Count / Size）: Evaluated based on the model size in MB
  • $m$（Model Parameter Count / Size）: Evaluated based on the model size in MB

Rules

  1. The competition consists of two stages: the Preliminary Round and the Finals. The list of finalists advancing to the Finals will be announced after the Preliminary Round concludes.
  2. All team members are required to attend the designated technical workshops for each competition stage in full, including both the Preliminary Round Technical Workshop and the Final Round Technical Workshop (held either on-site or online, as announced by the organizer). Full attendance by all team members is mandatory for each workshop. Teams that fail to ensure the full participation of all members will be disqualified from the competition. 
  3. During the Preliminary Round, the maximum number of daily uploads is 3. The final upload of the day will serve as your score for that day. In the event of identical scores, rankings will be determined based on the submission time; earlier submissions will rank higher.
  4. Participating teams are strictly prohibited from using multiple accounts, manually labeling test data, attacking the leaderboard system, or using any other unfair methods to manipulate the competition results. Violators will be disqualified from the competition and prize eligibility.
  5. Participants are allowed to use open-source data augmentation datasets and open-source pre-trained models, provided that all resources used are publicly available under open-source licenses. The sources of these data augmentation datasets and pre-trained models must be explained in a document and submitted along with the results. Participants may also utilize the baseline model provided by our institute as their pre-trained model.
  6. The data, technology, and source code used in the competition entries must be original to the participating team or legally authorized. If any infringement of intellectual property rights is verified by the organizer, the team will be disqualified from the competition and prizes, and the team shall bear all relevant legal responsibilities.
  7. Finalists or winning teams must submit model weights, source code, execution instructions, and related documentation as requested by the organizer for verification and acceptance. The authorization of the submitted models is limited to academic, public welfare, and official uses by the organizer, and shall not be used for commercial activities.
  8. Participating teams must achieve an mAP@0.5 of 0.6 or higher in the Preliminary Round to qualify for the Finals. If fewer than 10 teams meet this standard, the Finals may be left unawarded or the format may be adjusted at the organizer's discretion.
  9. Training rules: Models must be trained using the original 34 source categories.​
  10. To ensure fairness, regular employees, contract personnel, and individuals assisting in the execution of this competition from the organizing and co-organizing institutions (including the National Academy of Marine Research, DIGITIMES, ITRI, and AWS) are not eligible to register.
  11. The organizer reserves the right to adjust the competition system, data release methods, evaluation rules, and announcement details.
  12. Matters not covered herein will be resolved through mutual consultation based on the situation at the time.
  13. By completing the registration, all participants are deemed to have read and agreed to abide by the competition regulations and all relevant rules. 
  14. In case of any dispute, the organizer reserves the right of final interpretation of the event.
  15. Please read the Supplementary Provisions for other competition regulations.

ITRI Logo
Topics
Topic List
Courses
CTSP
ASVDA
Partnership
Playground
About
About Us
Our Team
Privacy policy
FAQ
System Status
Contact

For further suggestions or questions
admin.AIdea@itri.org.tw

Facebook
繁體中文 English

Copyright © 2017–2026 Industrial Technology Research Institute（AIdea platform v1.7.10）