A company uses SageMaker image classification (ImageNetV2 CNN) but finds the model misses many less-common animal species. They collected 10,000 labeled images of rare species in S3 and plan to train using Pipe mode. Which combination of steps should the ML engineer take? Select two.

To address the issue of missing less-common animal species, the ML engineer should perform transfer learning by fine-tuning the existing model. This leverages the pre-trained knowledge of the ImageNetV2 CNN and adapts it to the new, rare species data, which is more efficient and effective than training from scratch, especially with a limited dataset of 10,000 images. For SageMaker's built-in image classification algorithm in Pipe mode, the input data format typically requires a .lst file that maps image paths to their corresponding labels. This file, uploaded to S3, allows the algorithm to efficiently access and process the training data. Choosing a ResNet architecture and training from scratch is inefficient and likely to perform poorly given only 10,000 images for a complex task. Using an Inception model without fine-tuning won't address the specific problem of rare species. While an augmented manifest file in JSON Lines format is used for some SageMaker services, the built-in image classification algorithm with Pipe mode primarily uses the .lst format for image classification tasks.