Transfer Learning and Mixup for Fine-Grained Few-Shot Fungi Classification

2025-07-10

AI Summary

Fine-grained visual categorization (FGVC) of fungi faces significant challenges due to minimal inter-class discriminability and substantial intra-class variability, exacerbated by scarce labeled data. Method: This work proposes a domain-adaptive Vision Transformer framework, developed for the FungiCLEF 2025 competition and evaluated on the FungiTastic Few-Shot dataset. It integrates three key components: (1) domain-specific pretraining on fungal imagery; (2) a structured prompting mechanism combining weighted sampling with text-guided alignment; and (3) a few-shot fine-tuning strategy leveraging Mixup augmentation and transfer learning. Contribution/Results: The method consistently improves discriminative performance under both zero-shot and few-shot settings. On the FungiCLEF 2025 private test set, the final model ranks 35th out of 74 submissions—substantially outperforming baseline approaches. This demonstrates the efficacy of domain-adaptive modeling and multi-source information fusion for fungal FGVC.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Multi-modal Vision
• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Web Mining and Content Analysis: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Accurate identification of fungi species presents a unique challenge in computer vision due to fine-grained inter-species variation and high intra-species variation. This paper presents our approach for the FungiCLEF 2025 competition, which focuses on few-shot fine-grained visual categorization (FGVC) using the FungiTastic Few-Shot dataset. Our team (DS@GT) experimented with multiple vision transformer models, data augmentation, weighted sampling, and incorporating textual information. We also explored generative AI models for zero-shot classification using structured prompting but found them to significantly underperform relative to vision-based models. Our final model outperformed both competition baselines and highlighted the effectiveness of domain specific pretraining and balanced sampling strategies. Our approach ranked 35/74 on the private test set in post-completion evaluation, this suggests additional work can be done on metadata selection and domain-adapted multi-modal learning. Our code is available at https://github.com/dsgt-arc/fungiclef-2025.

Problem

• Address fine-grained fungi classification with few-shot learning
• Improve accuracy despite high inter-species variation
• Explore multi-modal learning with vision and textual data

Innovation

• Utilized vision transformer models
• Applied data augmentation techniques
• Incorporated domain-specific pretraining strategies