AGFD: ODIR-5K Few-Shot Retinal Disease Classification Results

78.7% accuracy and 76.9% macro-F1 scores were achieved in the 5-shot setting by the proposed Adaptive Gating and Focal Debiasing (AGFD) model on the ODIR-5K dataset. The supplementary file, authored by Stewart Muchuchuti and last updated in May 2026, contains results demonstrating performance improvements, especially for minority disease classes like Glaucoma and Cataract. The 20.5 KB document is shared under a CC-BY-4.0 license on figshare.

Use Cases

• Benchmarking few-shot learning methods based on the reported accuracy and F1 scores for retinal disease classification.
• Analyzing model performance on underrepresented classes based on the described +11–15 percentage-point F1 gains for Glaucoma, Cataract, Hypertension, and Other.
• Studying adaptive feature fusion mechanisms based on the described Dynamic Attention Gating module and its weighting analysis for global vs. local features.

Strengths

• Performance metrics are explicitly provided, including 78.7% accuracy and 76.9% macro-F1 for the 5-shot setting.
• The method shows specific, quantified improvements for minority classes, with F1 gains of 11–15 percentage points.
• The document is shared under a permissive CC-BY-4.0 license, facilitating reuse.

Limitations

• The dataset is a 20.5 KB supplementary document (DOCX), indicating a very limited scope focused on results, not raw image data.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for direct data analysis.

Provenance

Source

figshare

Collection Method

Likely contains experimental results from model evaluation on the ODIR-5K dataset.

Freshness

Last updated 2026-05-26 04:39:12; freshness should be verified.