Hybrid Fine-Tuning Strategy for Few-Shot Classification

Figure 1

Main idea and flowchart of the proposed HFT method for FSC. HFT performs the fine-tuning process based on the pretrained model. It includes an FSLDA module and an AFT module. FSLDA constructs the optimal linear classifier under the few-shot conditions to get the FSLDA model. AFT executes adaptive epoch learning and model performance evaluation using the validation classes of the base dataset to obtain the hybrid fine-tuning strategy, which is finally adopted for fine-tuning the pretrained model using the target dataset to get the HFT model.