An e-commerce company classifies product images by product line. A computer vision model trained with an image-classification algorithm has low accuracy on new products. Images are uniform size and stored in cloud object storage. The company needs faster improvement to deploy for new products. Which steps would raise accuracy? (Choose three.)

Augmenting existing images increases the diversity and quantity of training data, helping the model generalize better to new, unseen product images and reducing overfitting. Pixel normalization and image scaling are crucial preprocessing steps that standardize pixel values and image dimensions, which improves model training stability and performance. Checking for and addressing class imbalance ensures the model doesn't become biased towards majority classes, leading to better accuracy across all product categories. Training a semantic segmentation model is for pixel-level classification, not overall image classification, and wouldn't directly address low classification accuracy. Using a general-purpose image recognition API for automatic labeling might introduce errors or fail to capture specific product nuances, potentially degrading accuracy. A custom-labels image-recognition service builds a new model, but without addressing data quality or quantity issues, it might not offer faster or sufficient improvement.