Workshop Competition
FishNet Challenge
A text-guided open-set fine-grained fish classification benchmark for multimodal species recognition under naturally long-tailed distributions.
Overview
Text-guided open-set fine-grained fish classification
FishNet advances multimodal species recognition for marine ecology. Participants will develop models that integrate visual data and expert text descriptions to classify marine species.
Models are trained on a set of seen classes with textual metadata and evaluated on both held-out instances of seen classes and entirely novel unseen species.
Details
Competition information
Task description
Build models for fine-grained marine species classification using images and expert text descriptions.
Open-set recognition
Systems will be evaluated on known species as well as novel unseen species that do not appear in the training set.
Evaluation
Submissions are ranked using a unified metric that balances in-set and out-of-set detection, fine-grained accuracy on known species, and zero-shot generalization to unseen species.
Participation
Challenge platform, data access, and submission-format details will be posted here.
Timeline
Important dates
Competition launch
May 31, 2026
Final submission deadline
August 31, 2026
Winners announced
September 8-9, 2026, at the workshop