Label-free deep learning-based species classification of bacteria imaged by phase-contrast microscopy
Fig 9
Confusion matrices of the models trained once from scratch using the same random seed. All models were trained and tested at full resolution. The Video ResNet “R(2+1)D” used all 32 frames during both the training and testing phases. The single-frame classifiers ResNet-26 and ViT-B/8 trained on all frames in the time-lapse but could only access the first frame during test time.