Self-Detection Fine-Tuning: A Framework for Enhancing The Performance of Learning-Based Deblurring Models on UAV Data
Weitao Yue, Xiaowei Zhao · 2024
Affected by unstable disturbances, UAV-based visual data inevitably contains partially blurred segments, consequently reducing the overall effectiveness of the data. Additionally, considering the pronounced regional characteristics of UAV visual data, existing deep learning-based deblurring algorithms face challenges due to the absence of local UAV datasets. To bridge this gap, this paper proposes the Self-Detection Fine-Tuning (SDFT) framework. SDFT only takes a pre-trained deblurring neural network and a blend of clear and blurred UAV data as inputs. This framework filters input data through pre-trained neural networks, obtaining datasets suitable for neural network fine-tuning, thus eliminating the need for any additional datasets or algorithms. Experimental results validate the effectiveness of the SDFT framework in handling UAV-acquired visual data.