Are you skilled in machine learning models? Win up to 20,000 NOK!

Submission deadline: January 31, 2025

Updates:

December 21, 2024: An eagle-eyed contestant found that many of the ring codes in the RF dataset were wrong. This error arose from one of the Excel sheets used to annotate all the codes. About 15 % of the codes in the original dataset were wrong. Luckily, the mistake was not hard to fix. The corrected csv file has now been added to the download link below. The corrected file can be downloaded seperatly and should replace 'datasets/rf/ringcodes.csv'.

For many years, we have marked gulls with color rings throughout Norway – perhaps you have seen them in Nygårdsparken or Byparken? Now we need your help to take the project further!

We are looking for a new and efficient machine learning model that can read the codes on the color rings in our images. We have set up cameras in several locations and have already collected a large dataset of images, but we need a model that can automatically identify and read the codes on the rings. Can you solve the challenge?

We are looking for the most accurate and efficient solution. The model must be able to handle images from various environments and detect codes on rings it has never seen before. We have already developed a YOLO model that can be used to detect the rings themselves, but we need a model that can also automatically read the codes. This model will help NINA in their efforts to monitor seabird populations throughout Norway.

Prize Structure:

  • 1st place: 20,000 NOK
  • 2nd place: 8,000 NOK
  • 3rd place: 3,000 NOK
*paid as wages from the Norwegian Institute for Nature Research and thus subject to taxation

Participants will have access to a training dataset with a total of 35,354 images, where the codes are linked to the image files. The task is to develop a model that can reliably read codes from new images. Example data looks like this:

Example Images:

Bounding Boxes:

We have run the model to detect rings in the images. These bounding boxes are included in the download.

CSV File:

 filename|code|color
 20240603-160036(-178.66,-6.66,9231.0).jpg|J411H|NW
 20240607-070124(119.81,-2.21,10909.0).jpg|JC76T|NW
 20240608-100057(158.44,-4.73,8107.0).jpg|JE24X|NW
 20240712-220514(122.49,-14.15,5250.0).jpg|J808V|NW
 20240803-110159(117.49,-2.44,10538.0).jpg|JC77T|NW
 20240408-145102.064779(162.29,-4.57,9209.0).jpg|JE24X|NW
 20240408-145142.722154(103.96,-3.69,10150.0).jpg|JE38C|NW
 20240408-145459.652977(162.36,-4.64,7886.0).jpg|JE24X|NW

How to Participate:

 1)  Please register below so we have an idea how many is attending :)

 2)  Download all the images and the CSV file from Sikt FileSender (19.4 GB).
   

 3)  Train a model that can read the codes on the rings in the images.

 4)  Submit your model along with documentation and any results from the test data to sindre.molvarsmyr@nina.no.

The models will be evaluated based on their performance on a test dataset that is not available to participants.

Register for the Competition

You need to register to participate in the competition. Stay updated on the competition and receive notifications if important information arises by registering for the competition. We only send important updates about this competition, and email addresses will be deleted after the competition is completed.

Restrictions and Rules:

  • Scripts for training and prediction must be included in the submission.
  • The model must be able to run on a regular PC with Ubuntu.
  • The model must be compatible with Python 3.8 or newer.
  • Pre-trained models are allowed.
  • The model must be able to read the codes on the rings in the images and produce a CSV file with the filename, code, and color of the rings in the image.
  • Images where its detected two or more rings must have each ring on a separate line in the CSV file.
  • Group work is allowed (but you have to share the price).
  • To qualify to win a prize, the model must be able to correctly read at least 50% the codes in the test dataset.

Submission:

The submission deadline is January 31, 2025. Prizes will be awarded to the top three models, and the winners will be contacted and published here.

Don’t miss this opportunity to showcase your machine learning skills and contribute to an exciting nature research project – sign up now!

Evaluation:

The models will be tested on a test dataset that is not available to participants. They will be evaluated based on their accuracy in reading entire codes on the color rings in the test images. Only completely correct codes with the right predicted color will count as matches.

Questions?

If you have questions about the competition or the dataset, contact Sindre Molværsmyr at sindre.molvarsmyr@nina.no.

Video of the YOLO model running on an AXIS surveillance camera to detect and capture images of rings