Label-free deep learning-based species classification of bacteria imaged by phase-contrast microscopy
Fig 7
Video ResNet performance at varying number of frames.
Results from training video classification “R(2+1)D”-networks using various numbers of frames. Ordered and randomly shuffled time-lapses were compared for a fixed number of frames. The scatter plot shows class-average accuracy for each classifier, the line graph shows average class-average accuracy over all classifiers retrained with their respective train/test split. Lines are not for interpolation or statistical inference but are added as a visual guide to track the mean values across the ordinal scale.