Update README.md

README.md

@@ -4,6 +4,6 @@ language:
 - en
 pipeline_tag: image-classification
 ---
-Our approach is to design a visual recognition model that considers both the target object and its surrounding environment. To further enhance prediction consistency, we apply a custom loss function that reflects the hierarchical structure of the target categories.
+This is a visual recognition model designed for fine-grained classification of marine species. It incorporates both the target object and its surrounding marine environment, enabling more accurate interpretation of visually ambiguous underwater animals. The model leverages multi-scale contextual inputs to capture interactions between the object and its habitat. In addition, a custom training objective is used to reflect the hierarchical structure of marine biological taxonomy.
 
 While we employed standard classification accuracy for training, the primary evaluation metric was hierarchical distance, defined by the number of hops between the predicted label and the ground truth in the taxonomy tree. The final model attained a public distance score of 1.62 and a private distance score of 1.45 on the official evaluation leaderboard.