Mastering Computer Vision with TensorFlow 2.x

Analyzing and storing data

First, we will start by analyzing and storing the data. Here, we are constructing a furniture model with three different classes: bed, chair, and sofa. Our directory structure is as follows. Each image is a color image of size 224 x 224.

Furniture_images:

• train (2,700 images)
  • bed (900 images)
  • chair (900 images)
  • sofa (900 images)

• val (300 images)
  • bed (100 images)
  • chair (100 images)
  • sofa (100 images)

Note the number of images is just an example; it is unique in every situation. The important thing to note is, for good detection, we need about 1,000 images per class and a train and validation split of 90%:10%.