Forecast-PEFT: Parameter-Efficient Fine-Tuning for Pre-trained Motion Forecasting Models

Forecast-PEFT introduces a parameter-efficient fine-tuning framework for motion forecasting that utilizes specialized prompts and adapters to minimize retraining overhead. The method addresses the misalignment between masked reconstruction pre-training and downstream trajectory prediction tasks by preserving most pre-trained parameters.