We have been downloading Inception v3 weights and then loading the weight into an instance of the InceptionV3 model. Would we get the same results if we included the weights if we wanted to? are the weights we download the same as if we include the weights?
Setting weights means that the model parameters are set to the provided values. You still might not get the same results due to the following reasons:
- Shuffling of data.
- If internal state of optimizer is different across platforms when you start training.
You don’t have to worry about that level of detail though. By downloading the weights, you are solving a bigger problem i.e. training a model with fewer data points since the weights were obtained by training the model on a much bigger dataset.