Comparative study of adapting pre-trained models for driving behavior video captioning
The study compared full fine-tuning, LoRA adaptation, and prompt engineering to adapt pre-trained models for driving behavior video captioning using BDD100K and BDDDX datasets. Full fine-tuning achieved the best performance across all metrics, though it is computationally expensive. The findings suggest that while efficient methods exist, deep parameter adjustment offers superior results for this complex task.