Figure 6: The training curves of fine-tuning CNNs on Scene 15 dataset and MIT 67 dataset with the pretrained models.