LoRA-Based Diffusion Data Augmentation for Underwater Garbage Detection: An Empirical Study with YOLOv8s
Abstract
Abstract Underwater garbage images are difficult to acquire and are often affected by light absorption, scattering, suspended particles, and insufficient illumination. These factors commonly lead to color distortion, blurring, low contrast, and noise, making deep learning-based detection models vulnerable to limited data availability, class imbalance, and scene variation. To investigate the feasibility of generative data augmentation for underwater garbage detection, this study develops a synthetic data augmentation workflow that integrates Stable Diffusion v1.5, LoRA-based class-specific appearance adaptation, and ControlNet-based spatial conditioning. YOLOv8s is adopted as the downstream detector to empirically evaluate the effect of different synthetic data strategies. To reduce the risk of data leakage, the real-image dataset was first divided at the image level into training, validation, and test sets before any generative model adaptation or synthetic image generation was performed. The images used for LoRA adaptation, the ControlNet spatial hint maps, and all synthetic images were derived only from the training set. The generated images were added only to the YOLOv8s training set, while the validation and test sets remained composed exclusively of real images. The experimental results show that adding LoRA-adapted synthetic data improved the Precision from 83.3% to 85.0%, Recall from 80.2% to 81.8%, [email protected] from 83.9% to 86.7%, and [email protected]:0.95 from 55.9% to 57.2%. In…
Read the paper
Related studies
- Learning from First-Generation Qualitative Approaches in the IS Discipline: An Evolutionary View and Some Implications for Authors and Evaluators (PART 1/2)
- Commentary on the clinical and preclinical dosage limits of interstitially administered magnetic fluids for therapeutic hyperthermia based on current practice and efficacy models
- Prosthesis: New Technological Opportunities and Innovative Biomedical Devices
- The history and future of neural modeling for cochlear implants
- Ryanodine receptors: waking up from refractoriness
- Novel Transgenic Rabbit Model Sheds Light on the Puzzling Role of Matrix Metalloproteinase-12 in Atherosclerosis
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.