Do you require 10 or 50 min labeled images?

Question

Hi,

I'm unsure:

Do you need to label a minimum of 10 or of 50 images for labeling? Or is there a difference between the labeling for images and the labeling for object detection? The text doesn't really clarify:

min 10: https://learn.microsoft.com/en-us/training/modules/create-labeled-dataset-using-azure-machine-learning-data-labeling-tools/3-label-images-azure-machine-learning-data-labeling-tools

min 50: https://learn.microsoft.com/en-us/training/modules/create-labeled-dataset-using-azure-machine-learning-data-labeling-tools/4-export-labeled-azure-machine-learning-dataset

Bye

Michael

Answer 1

Hi Michael Schmidt,

Welcome to Microsoft Q&A! Thanks for posting the question.

In Azure Machine Learning, it is recommended to label at least 10-15 samples per label to prepare a dataset effectively for initial training. The more diverse and numerous the labeled samples, the better the model will perform. For training the model, a larger dataset is necessary, with at least 50 labeled images per class to ensure robust performance, especially for complex tasks. For more info, please refer the document Labeling images and text documents - Azure Machine Learning.

For object detection tasks, the requirement is typically higher. More labeled images are needed to achieve good accuracy, especially if the objects vary significantly in appearance, size, or context. This ensures that the model can effectively identify and localize objects across different scenarios.

There is also a distinction between image classification and object detection. Image classification assigns a single label to the entire image, identifying the main subject (e.g., labeling an image as "cat"). In contrast, object detection identifies and locates multiple objects within the same image, providing class labels and bounding boxes. For instance, in a street scene, object detection can identify and locate objects such as "car," "pedestrian," and "traffic light." This distinction often requires more labeled images for object detection due to its complexity.

I hope you understand. And, if you have any further query do let us know.

---

If this answers your query, do click Accept Answer and Yes for was this answer helpful.

Thank you!