FungiCLEF — Next Steps, Final Leaderboard, and Recent Changes.
Dear All,
As the competition deadline is approaching, I would like to take this opportunity to let you know about the final evaluation, submission process, and a few changes in the public leaderboard.
Number of Submissions: The number of submissions per day was increased to 5 a day. However, you don't have to overfit to the leaderboard - the test set is entirely different. Thus, your methods will perform differently.
Submission Process: The final evaluation will be done on our server. We expect you to:
Upload your model with a max size of 1GB to any private repository, e.g., HuggingFace Hub, TensorFlow Hub, and Pytorch Hub.
Send us an email before the given deadline: (i) TeamName submitted to the provided registration form, (ii) TeamName on HuggingFace, and (iii) access to a private GitHub repo (users: picekl and chamidullinr) with an inference script and Docker file / Conda environment that will perform the evaluation.
We will provide a baseline repository by the end of the week!
Final Leaderboard: The Private Leaderboard will display the competition scores once the competition evaluation process ends. Everyone who will not provide a GitHub Inference code will be removed from both leaderboards.
Test set data: The test set images will never be provided in order to prevent cheating. We expect you to validate your findings on the validation set. The metadata format in the test set is the same; therefore, the evaluation should be done quickly on our end.
Working Notes: As you might know, this competition takes part in two workshops [FGVC + LifeCLEF] / conferences [CVPR + CLEF], and it is required that participants write a scientific report / working notes(WN) article about their work to sustain reproducibility.
Teams not providing the WN paper will be removed from the leaderboard.
Post-competition evaluation: The post-competition evaluation on the test data will be enabled for your intermediate evaluation, but we will perform it once a week for all new submissions (a limited number for each team).
PS: We have noticed few teams are creating multiple accounts to overcome the strict one-submission-a-day rule. I would like to point out that this is against the rules! We know who you are. Please send me an email to prevent your immediate disqualification!
PS2: Keep your code private. Last year, cheating participants were sniping GitHub and using code from others.
IN CASE ANYTHING IS UNCLEAR, DO NOT HESITATE TO CONTACT ME.
@word2vector
,
@asdadasasd
,
@asdadadasd
,
@sdgag
,
@headacheIsHeadache
,
@asasdsad
,
@masie
,
@rubengv1999
, @samuelise,
@transfrom
,
@mrzave
,
@processorprocess
,
@latinorigin
Lukas
Hi, we have submitted the results on 5.17, and why can't we see the result on the leadboard?
