Fix Object Detection Model Evaluation

[KazuOnuki]

In this classification model doc, it explains that custom vision services are basically conducting k-fold cross validation internally.

On the other hand, object detection model doc did not mention that the k-fold cross validation would be carried out with a custom vision.

In the object detection model, the customer may misunderstand that they do not perform k-fold cross validation on object-detection model, so we would appreciate it if you could correct the description just in case.

---

Document Details

⚠ Do not edit this section. It is required for docs.microsoft.com ➟ GitHub issue linking.

• ID: 074ad73e-f611-d9e4-d612-0813148c9202
• Version Independent ID: dc8df697-6110-2fbb-c74c-da12bedbc5aa
• Content: Quickstart: Build an object detector with the Custom Vision website - Azure Cognitive Services
• Content Source: articles/cognitive-services/Custom-Vision-Service/get-started-build-detector.md
• Service: cognitive-services
• Sub-service: custom-vision
• GitHub Login: @PatrickFarley
• Microsoft Alias: pafarley

[GiftA-MSFT]

@KazuOnuki we'll review your feedback and get back to you shortly. Thanks.

[GiftA-MSFT]

@PatrickFarley can you please clarify the process used in object detection model for estimating performance? Thanks.

[KazuOnuki]

@GiftA-MSFT I have just asked the development department about the how to train customo-vision model, and they have given me feedback that they have changed from the k-fold-cross-validation method to a simple training-validation method around 2020, and that they are splitting the training : verification = 4 : 1 ratio. Therefore, would it be possible to add this information to the report?

[PatrickFarley]

@KazuOnuki Does this mean that 1/4 of the training data is used for verifying the model? Can you provide more info about how the score is calculated? I can remove the note about k-fold cross validation

[PatrickFarley]

I've removed the mentions of k-fold

please-close