Mastering Computer Vision with TensorFlow 2.x

Preparing images

Follow these steps to prepare your image:

1. Study how many classes you want to detect—in this example, we will consider the bed, chair, and sofa classes discussed in Chapter 6, Visual Search Using Transfer Learning.
2. Ensure that you have the same number of images per class.
3. Make sure your class name does not have white spaces in it; for example, use caesar_salad instead of caesar salad.
4. Collect at least 100 images per class to start the initial training (hence, complete steps 1 to 10 to develop the model) and then increase the number as you have more and more images. Ideally, 1,000 images is a good number for training.

5. Bulk resize all the images to 416 x 416—you can select options in the macOS preview pane, then select multiple ...