Training a custom Object Detection model
You can train an IBM Watson Object Detection custom model by using the Visual Recognition model builder in IBM Watson Studio.
Before you begin
Creating an Object detection model
To create an Object detection model you must associate a Visual Recognition service with your project.
Restriction: A Visual Recognition service instance may be associated with only one project at a time.
- From the Projects page in Watson Studio, create a new project or open an existing one.
- Click Add to project, and then select Object detection model. If a Visual Recognition service instance is not already associated with your project, follow the prompts to create a new service or associate an existing service. Note: A Visual Recognition service instance can be associated with only one project at a time.
- Enter a name for your model.
- Add .zip files to your model from the data panel. (If the data panel isn’t open, you can open it by clicking the Find and add data icon ().)
Label your images
- For each image in your training set, click an image thumbnail to open the image, and click the edit icon.
- Use the mouse to draw bounding boxes around each object in the image that you want to detect and assign a label for the object. The label identifies what the object is. The location identifies where it is in the image.
- Use your mouse cursor to draw a box around an object in the image. Tip: You can select a different color for your boxes from the tool palette if they are not easily visible.
- Specify a name for that object and click Add in the image pane to confirm the object name, position, and size.
- Repeat these steps for other objects in this image. In this example, brain tumors are identified in a series of MRI scn images.
- Press Done in the image pane when you finished labeling objects in the image.
- Repeat the labeling steps for other images in your collection.
- After labeling a representative sample of images, you can train the model to get some labeling assistance from Watson with the remaining images. Click Train Model.
- Wait for the training process to finish. You can leave the Object Detection model builder while the process runs. You can monitor the status of the training process by looking at the status of the model on the Assets page of your project.
- When the training is complete, click to view an untrained image in the collection, then click the Auto Label icon to view objects that Watson has detected in the image. Review the labels in the image pane and accept the correct ones.
- Continue the labeling process until all of the images are labeled, then click Train Model to repeat the training.