What's the difference between fine-tuning vgg16 and create a separated VA model?

[Yjuun]

Hi,

It seems to me there are two ways of building the emotion model:

1. Extract features as a vector for each image and inputs these training vectors into another model (i.e. neurons in layers: 3000, 1000, 1000) to output a VA score. It's the one you built.
2. Fine-tuned the VGG16 model by adding fc layers (i.e. 3000, 1000, 1000) and input with training raw images, while output VA score.

I wonder if it is possible to use the second way, and if so, would it achieve the same result as the first method? It just occurs to me that the second method may be easier to set up, am I right?

Thank you!!