Artificial Intelligence Value-Chain
Analysing the entire haute couture production of the last 20 years and based on the designer’s needs in terms of fabric type, dress type, dress size and colours proposing an unlimited number of suggestions through images generated by AI who learned the creation techniques through training on a set of 30,000 images.
Use the latest achievements in Deep Learning to create the models, not before creating a reference dataset of 30,000 images from the main fashion collections of the last 20 years.
- Training a GAN (Generative Adversarial Network) neural network to recognise images from catwalk photos of fashion collections.
- Magnification of the training set by data augmentation assisted by a DAG (Data Augmentation optimised for GAN) neural network.
- Segmentation of the clothes that have been recoloured separately using a CNN (Convolutional Neural Network) neural network and a Threshold Segmentation algorithm.
- Generation of the images by a second neural network (part of the GAN) trained thanks to the first neural network and able to discriminate the generation of the images by discarding those that do not fall within the defined parameters.
- Develop an application that incorporates the above and has a simple, intuitive and efficient user interface.
- Optimise the application to work on a specific device with sufficient computational capacity to make the entire creation process last a few seconds.
Textile / Clothing Industry
Available on the market