ODIR-5K: Retinal Disease Classification Performance Metrics

78.7% accuracy and 76.9% macro-F1 scores were achieved on the ODIR-5K dataset in a 5-shot setting using a novel Adaptive Gating and Focal Debiasing (AGFD) method. The research by Stewart Muchuchuti, published in 2026, demonstrates improvements over hybrid-attention baselines, particularly for minority disease classes like Glaucoma and Cataract. The dataset likely contains performance metrics and analysis supporting the proposed few-shot learning model for automated retinal disease screening.

Use Cases

• Benchmarking few-shot learning models for retinal disease classification based on reported accuracy and F1 scores.
• Analyzing model performance on underrepresented disease classes based on reported +11–15 percentage-point F1 gains.
• Studying adaptive feature fusion mechanisms for medical imaging based on the described Dynamic Attention Gating module.

Strengths

• Performance metrics are specific, reporting 78.7% accuracy and 76.9% macro-F1 in the 5-shot setting.
• The method shows clear improvements for minority classes, with F1 gains of 11–15 percentage points for Glaucoma, Cataract, Hypertension, and Other.

Limitations

• The dataset is a 493.5 KB PDF, suggesting it contains summary results and analysis rather than 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 training models.

Provenance

Source

figshare

Collection Method

Likely contains experimental results and analysis from research applying the AGFD method to the ODIR-5K dataset.

Freshness

Last updated 2026-05-26 04:39:13; freshness should be verified.