A retail company trained an ML model on Amazon SageMaker to detect damaged products from warehouse surveillance images. The training dataset was created using SageMaker Data Wrangler, which included images of damaged and undamaged products. During training and validation, the model achieved high accuracy; however, in production, the model’s performance is degraded due to differences in lighting conditions and image quality across various warehouse locations. The ML engineer needs to improve the model's accuracy to handle variations in image quality with the LEAST amount of time and effort. What do you recommend?