What should the Specialist do to initialize the model to re-train it with the custom data?
Initialize the model with random weights in all layers including the last fully connected layer.
Initialize the model with pre-trained weights in all layers and replace the last fully connected layer.
Initialize the model with random weights in all layers and replace the last fully connected layer.
Initialize the model with pre-trained weights in all layers including the last fully connected layer.
Explanations:
Initializing the model with random weights in all layers would not leverage the learned features from the pre-trained model, which is crucial in transfer learning for improving performance with limited custom data.
Initializing the model with pre-trained weights in all layers allows the model to utilize learned features from the original dataset, while replacing the last fully connected layer enables the model to adapt to the specific vehicle classification task.
Initializing the model with random weights in all layers discards the benefits of transfer learning, which relies on pre-trained weights to capture important features, making it less effective for the new task.
While initializing the model with pre-trained weights helps in leveraging prior knowledge, including the last fully connected layer may not be appropriate since the final layer needs to be adjusted for the specific number of vehicle classes in the custom dataset.