Fig 1.
Trachoma classification of selected field collected images, according to the WHO simplified system.
TF: trachomatous inflammation—follicular; TI: trachomatous inflammation—intense [16].
Table 1.
Distribution of clinical categories in our dataset.
Fig 2.
Sample image where eyelid is neither centered nor horizontally aligned.
Fig 3.
An illustrative example for various eyelid images in our procedural pipeline.
Fig 4.
Network architecture of multilayer perceptron-based pixel-level classifier.
Fig 5.
(a) 256 × 256 crop on the rotated image. Estimated (white) and randomly perturbed eyelid centers (green) are shown. (b) Resulting 128 × 128 grayscale ROI.
Fig 6.
Convolutional layer with zero-padding and a 3 × 3 filter followed by max pooling with a 2 × 2 block.
Table 2.
Architecture of our convolutional neural network classification model.
K denotes the number of filters in the first stage of the convolutional layers.
Table 3.
Validation scores on trained convolutional neural network models for TF and TI classification tasks.