Source: aicep Portugal Global
Portugal-based Fashable is here to revolutionize the fashion industry with AI. Unsustainable manufacturing, unsold inventories, and long production cycles are common issues in fashion. But according to Orlando Ribas Fernandes, Co-Founder and CEO of Fashable, they can all be solved with the help of technology. Using Azure Machine Learning, an enterprise-grade service for the end-to-end machine learning lifecycle, and PyTorch, an open-source machine learning framework, Fashable created an AI algorithm that can generate original clothing designs, helping fashion companies to meet customer demand, get to market faster, and reduce clothing waste.
A style explosion
The world loves fashion. Fashion is one of the fastest-growing, most lucrative and demanding industries, with high expectations of quick turnaround rates, creative designs, and an endless parade of new styles.
What’s behind this explosion in production? Some experts say “fast fashion,” the term used to describe the mass-manufacturing of runway styles quickly and cheaply. While a high-end designer might take months (or years) to design and produce a collection, fast fashion brands can do it in a fraction of the time and, thanks to inexpensive labor and materials, at a fraction of the cost.
An unstable trajectory
In one sense, fast fashion is good for customers, who can buy designer looks for less. But critics argue it has led to the rise of “throwaway culture,” with some studies estimating that the average consumer discards an article of clothing after only seven to ten wears.
So, what happens when clothing production goes up while lifecycle goes down? A growing landfill problem. In fact, every year the United States alone throws away 21.6 billion pounds of textile waste.
In a world increasingly concerned with sustainability, fashion designers need to find a way to meet demand while reducing waste.
Bringing outside change to an insider industry
Orlando Ribas Fernandes, Co-Founder and CEO of Fashable, is the first to admit he’s a fashion industry outsider. With a master’s degree in Artificial Intelligence and Intelligent Systems, Ribas Fernandes has long been obsessed with the potential of AI and machine learning to disrupt industries and create positive change in the world. “My background is technology, and I’ve been working in artificial intelligence research and development for 15 years. A few years ago, we launched a joint research initiative with Microsoft to build innovative new technologies. After we built our model, we realized fashion designers can use our AI technology to work directly with their customers, which, for us, is priceless. So, we wanted to bring a new perspective to the industry and show the power of AI.”
The Fashable AI application can create dozens of original AI-generated clothing designs in minutes, without the need for actual material. The next step was finding a platform to support it. “When we started working in our AI technology, one of our requirements was to have a platform and a deep learning framework that aligned with our vision and strategy. We wanted a platform that could give our team the required tools to explore new innovative frontiers. That was the case with PyTorch and Microsoft Azure Machine Learning,” says Ribas Fernandes.
With PyTorch, Fashable has an open source machine learning framework that is powerful, flexible, and fast, and offers a rich developer community. PyTorch on Microsoft Azure is optimized for deep-learning workloads without the need for tedious environmental set-ups. Azure Machine Learning delivers massive GPU support alongside native integration and interoperation, freeing the organization’s research team from having to set up complex environments for building and deploying image-based models. As Ribas Fernandes puts it: “We consume a lot of GPUs training our models, and we’re a small startup that doesn’t have the capability to make huge upfront investments on those machines, so it was super important to find something that could support GPU usage in a cost-effective way. Azure helps us train our models without making those types of investments and helps our research team work faster.” Fashable reduced training costs by using low-priority virtual machines within Azure Machine Learning studio. The combination of Azure Machine Learning and PyTorch also provides an environment for accelerating and managing the lifecycles of Fashable’s machine learning projects with its MLOps capabilities.
Fashable also benefited from running their model with Distributed Data-Parallel implementation with Azure Container for PyTorch (ACPT). “We were able to successfully run the model with multi-modal nodes and get the results effectively and efficiently. When using an environment other than ACPT, we were not able to achieve that,” says Ribas Fernandes.Sustainable fast fashion
Creating a new fashion collection takes lots of time, money, and material. Fashion trends are unpredictable, making it hard for designers to know if their inventory is going to sell.
Fashable removes much of the labor and guesswork. Its AI is composed of different neural networks that ingest data from multiple sources like social media or commerce and retail sites to learn about trends, styles, and clothing types. The models are constantly learning what’s in fashion versus what’s out and can visually augment the digital design in real-time, like shortening the sleeves of a dress or changing a pattern from stripes to polka dots.
Designers can take these creations to social media to A/B test directly with customers, helping them gauge interest and forecast demand before going into production. Where it used to take months to get a new collection from design to department store, with Fashable it now takes minutes. Designers can create innovative pieces, market them directly to customers, and forecast demand without wasting a single scrap of fabric.
An intelligent future
Even though Fashable is using its underlying technology for fashion, Ribas Fernandes says the organization is starting to apply it to other use cases like healthcare, manufacturing, and the metaverse.
In healthcare, Fashable’s AI can be used to assist with things like lung cancer detection. Fashable’s model can read x-ray scans of both healthy and diseased lungs to learn to recognize signs of cancer and other illnesses. Ribas Fernandes believes AI can help improve the quality and accuracy of medical diagnoses, helping patients to receive medical treatment earlier in their illness.
According to Ribas Fernandes, the future of artificial intelligence combined with human ingenuity is bright. “I love AI. And I truly believe that when it’s used for good, it can help save lives and change the world.”