DUQ FedAvg Project
Home DUQ FedAvg Project Federated Learning CNN Backbones U-Net Models Dual-UQ

Parasite Classifier Application

Client-side application of federated parasite classification using DenseNet121, ResNet50 and VGG16 with uncertainty-aware aggregation.

CNN Model Selection

Choose which CNN backbones to include in this run.

Deep residual network, strong for general image features.
Efficient feature reuse, suited to medical imaging.
Classic CNN baseline, useful for comparison.
When DenseNet121 is selected, the Dual_UQ federated method is highlighted as the leading strategy in the performance plots.
Select Server Location
Federated Nodes

Visualisation of the central server in Nairobi and participating county nodes used in this run.

Nairobi (Central)
Kilifi
Murang’a
Kirinyaga
Kisumu
Upload Microscopy Images

Select one or more stool microscopy images to simulate the inference pipeline.

Model Performance

ROC and Precision–Recall curves are generated to mimic the behaviour of several federated optimisation methods. When DenseNet121 is active, Dual_UQ (dark red) appears as the leading method.

ROC Curve
Precision–Recall Curve
County Heatmap

Colour intensity reflects activity or data volume per county for this run.