Fig. 4
From: BAGLS, a multihospital Benchmark for Automatic Glottis Segmentation
Evaluation of model performance using the Intersection over the Union (IoU). (a) Cumulative distribution of the IoU across validation (magenta) and test (blue) set. Shaded error shows 95% confidence interval of bootstrapped distributions. (b) Distribution of IoUs against segmented area in the ground truth. Left: validation set; right: test set. (c) Example images and segmentations for IoUs close to 0.25, 0.5 and 0.75. Intersection of segmented pixels in the ground truth and prediction in green. Blue and red pixels were classified as glottis only in the ground truth and only in the prediction, respectively. (d) An example video from the BAGLS dataset (0.mp4, same subset as in Fig. 3c) segmented (orange overlay) using the trained model with respective glottal area waveform (sum of segmented pixels over time).
