Generating watch models based on GAN

Description

How to use AI to create customized images of luxury watches
In recent years, new neural networks called "generative adversarial networks" - GANs for short - have been able to penetrate areas that were previously reserved exclusively for humans. Creativity or art is not typically considered a domain of computers. However, since the resurgence of GANs in science in 2014, generative models have increasingly found applications in this field.

In short, the styleGAN architecture allows you to control the style of generated examples within the image synthesis network. This means that you can customize the high-level styles (w) of an image by applying different vectors from the W space. Additionally, you can transfer the style from one generated image to another. These styles are mapped to the LOD (level of detail) subnets of the generator, which means that the effect of these styles ranges from coarse to fine.

Mixing Styles

Now let's look at transferring the style from one generated image to another.

Pose and Shape

The pose of the watch is represented by the first four dimensions of the hidden style vector. If we adjust the vector, we can change the position of the generated image, such as rotating the watch along the axes, as well as changing the shape of the watch itself. Even more interesting is the transfer of pose and shape, where we can take the pose of any watch and transfer it to our target watch.

Dials and numbers

The style of dials, sub dials and numbers is also represented by a hidden style vector (dimensions 4-8), and changing it has the same effect as above. But we can clearly see that the changes are more about the details of the watch, rather than the shape and pose.

Colour and appearance

The lighting setup, background and colour of the watch are encoded in the last dimensions of the style vector (dimensions 8–18). Here we can control the colour and material of the watch, as well as the background and lighting setup of the image.

If you are interested in this neural network and it can help you solve your business and other technical problems, please send us an email: info@ai4b.org