Figure 1.
Transferring parameters of convolutional neural network (CNN). First, network is trained on source task (ImageNet classification, top row) with large amount of available labeled images. Then, convolutional layers (C1-C5) are transferred to target task(brain tumour and encephalitis classification, bottom row). We remove original fully connected layers (FC6-FC8) and add adaptation layer (FCa-FCc). Last, we fine-tune new model on labeled data of target task.
