How to append images, annotations and predictions to an existing dataset.

🤔 Imagine you have an existing dataset and you wish to add new data. This walkthrough will guide you through the steps you need to follow to accomplish this task

Appending Images


Before you start: Parent folder location

We assume that all your images in the dataset share the same parent folder, which is defined during dataset creation. This is important because our image keys, URLs, paths, etc., are then all based on the relative image paths.

☎️ If this does not align with your use case, please reach out to our sales team for custom location support tailored to Enterprise needs.

To append new images to an existing dataset, follow these steps:

  1. Place the new images in the same parent folder as the original dataset.
    1. Note that they may have any folder structure beyond that; for instance, you can create a new subfolder inside the parent folder and place the images there 📂.
    2. 👉 For example: you can have three subfolders named train, val and test under a common dataset.
  2. Re-run the dataset ingest endpoint.
  3. This process will automatically 🔄 scan the parent directory and add any new images, while ignoring existing ones.

Appending Annotations


Before you start: Annotations file

We assume that the user is providing a file with the new annotations only. We will address other scenarios, such as files with additional images to be ignored, at a later stage.

To append a new annotations file:

  1. Utilize the "Upload Dataset Annotations" endpoint for the new file.
  2. Subsequently, invoke the ingestion endpoint for the dataset annotations.

Appending Predictions

To append more predictions:

  1. Use the "Upload Model Predictions" endpoint for the new file.
  2. Call the ingestion endpoint for model predictions.