Can a pre-trained model be called?

[aidongmandexiaowowo]

Could you please enlighten me on whether it's possible to leverage pre-trained models for training, given that training from scratch results in subpar network performance? Furthermore, I'm curious to know what constitutes an adequate amount of data, as the Oxford-Paris dataset seems too vast, and I'm wondering if a smaller dataset can still produce a satisfactory outcome