About the training with the 91 image dataset and fine-tuning with the General 100 dataset

[johnniewalker3]

One last question Saafke, when we train a network from scratch we train it with the 91-images datasets.When we have to fine-tune it we have to train the network with general 100 dataset and with 91 images dataset or only with general-100 dataset?

[Saafke]

In the original paper and my code, we only use the general-100 dataset for finetuning. During the finetuning stage, the learning-rate is halfed (compared to the training stage).

But you raise a valid point, you could actually use both datasets during finetuning! Why don't you try out, and see which strategy works better? :)