Online fashion outlet Otrium and software developer Pixyle.ai are working together to scale the operation with automation and enriched product data. For example: with the auto tagging software of Pixyle AI, E-commerce companies are provided with a more intuitive and effortless search experience using AI and computer vision technology.
Svetlana Kodumova, founder of Pixyle.AI founded the company because she felt that e-commerce search was letting shoppers down. “Product search is a frustrating experience for many people, when they are not able to find what they were looking for. Websites are continuing delivering irrelevant search results and are failing to meet the expectations of the shopper, and this is costing the industry enormous amounts of money. We believe that visual AI is the answer to this growing frustration of online shoppers.”
Pixlye.AI helps e-commerce companies to provide a more trouble-free search experience by using AI and computer vision technology. The software enables them to improve product discovery by helping their shoppers find exactly what they’re looking for at a faster rate than before. They offer three different AI-solutions:
- Automatic tagging which extracts rich product attributes from images automatically.
- Visual search which enables shoppers to find the product they’re looking for by uploading a photo.
- Similar recommendations with suggests items like the product that the shopper is looking at.
The technology behind everything
According to Svetlana the underlying technology behind their solutions is computer vision, a subfield of deep learning and AI. It enables computers to get insights and interpret visual information from images and videos.
“Our AI-models perform at the highest level with an accuracy of more than 90%. At the start we created a data-centric AI-strategy with a large focus in building quality training data for our AI-models. We’ve set up our own in-house annotation team that manually labels all the data that goes into the training dataset and implemented strict data quality control processes”, thus Svetlana.
Pixyle delivers highly accurate and detailed product attribution for fashion ecommerce by having an extreme focus on good data. The company manually labels images in-house, with strict quality control. It has built a proprietary dataset of over a million carefully labeled images and it works with an extensive fashion-specific taxonomy of more than 1,000 attributes that describe all possible product characteristics that you can think of.
The system that can understand fashion in images works as follows:
- The system finds the location of each fashion item within the image,
- Then it determines the category the fashion items belong to, for example “dress” or “sneaker”.
- At the end the system digs deeper into the image on a more granular level. It detects and classifies each unique attribute, for example “color” or “pattern”.
Here you can find a demo of the system.
The collaboration with Otrium
Pixyle.ai focused specifically on fashion as they wanted to provide the highest level of detail with the highest accuracy in predictions possible. The collaboration started at the end of 2020. Bradley King, Head of Product and Operations at Otrium, saw that their product catalogue was growing massively, and they needed to tackle two major pain points and opportunities. “We needed to improve the efficiency around tagging and processing product attributes as this was done manually by internal team. Also, the product discovery for our customers wasn’t provided good enough. With Pixyle we were able to break down the process into smaller iterations to ensure fast implementation and feedback loops for further learnings and improvements.”
Since the collaboration Otrium has had a 65% improvement in productivity and efficiency by automating color detection through the AI-solution for their inbound logistic team. The improved product discovery for the customers by enabling different filters like style and patterns are a huge advantage in making the search for the right clothing more efficient. “Efficiency advantages for implementing new filters as our products are tagged with many product attributes and product discovery are huge advantages for Otrium since we’ve started the collaboration with Pixlye.ai”, thus Bradley.
Since the start of the collaboration, Pixyle has become Otrium’s AI-partner, deepening the relationship and helping them on more cases. “Together we’ve developed a new product called “Shot type detection”. This solution determines if the image type is a “Packshot”, a stationary image, or if it contains a “Model” – a person wearing fashion items. For this purpose, we trained a completely new AI-model, one that can understand the shot type of an image. Otrium is now able to automatically display the photo with a model as the first one on the website”, thus Svetlana.
Mission for the future
According to Svetlana, Pixyle is on a mission to change the way people discover products online by making all digital content shoppable, leveraging the power of visual AI. “We aim to help all E-commerce companies build better product search by offering a complete suite of content enrichment. There are still so many problems within this market, and we want to be the go-to AI-partner with many solutions that can help our customers.”
Bradley King is happy with the collaboration with Pixyle. For the future there are many plans to make the search for the right clothing for the customers even better. “We want to add a validation tool back to Pixyle, leverage attributes as product descriptions and use category attributes as pre-categorization so that we can minimize manual effort of assigning a category to products.”