Use cases

We are working in collaboration with Stefy in order to create her collection of NFT’s. The goal is to represent her best moments in competitions, best tricks, which will provide to their fans access to experiences. In the first stage we will publish just a subset of the collectibles in OpenSea with hidden features. In the second stage we will reveal the features in a random way. We will publish the smart contract in the Polygon and Ethereum blockchains. The profit mechanism will be based on royalties.

The supplied solution began with the development of a Data Lake in Azure, which was then structured and processed using Databricks and Pyspark. The next step was to integrate all the numerous data sources from the club data providers and clean and structure them to create an initial data lake. Following that, the next phase was to define how the various departments wanted to visualize and operate with the data, as well as to prepare all the tables and views. After we have completed the previous phase, we will establish the analytical model and the relationships in Power BI between all the dimensions of data in order to have an automatic model capable of creating graphics and visualizations in an agile and optimal manner. The next phase was to define all the panels and graphics that the users wanted to visualize and work with. The solution is totally customizable and may provide both an overview and specific information on everything that is going on in a company.

The partnership began with a definition of the most typical ways criminals steal merchandise from supermarkets. Based on it, we began working on the databases that would be used to train and evaluate the models and applications.

The approach presented was based on computer vision and deep learning techniques. The solution’s core consists in employing deep learning models to extract information about objects and important areas on the person’s body from video cameras placed along the shelves. We then used this data from tracking methods and traditional computer vision algorithms to extract temporal information from the items. Finally, all this spatial and temporal data is evaluated by hand-crafted algorithms to detect suspect behavior. The program was placed on an AWS cloud service and made available via an API for consumption by the front-end application, which displayed the camera image and the results of the algorithms.

In addition, to localize those empty spots, we construct an object identification model based on a cutting-edge neural network. We also created the database used to train the model in partnership with Accenture.

We collaborate with them in the development of these applications based on classical and state of the art computer vision algorithms. For some applications like tampering we use classical techniques like optical flow or background subtraction, meanwhile for more complex tasks such as mask detection we take advantage of more recent convolution network models. The models were deployed as edge applications, e.g. they were running on the devices themselves, sending the alerts to its event hub deployed on the cloud.

We define with ByDeluxe a set of useful tools, centered in computer vision VX, that will highly speed up the post-production process in a couple of Netflix series that they were working on. The process required to develop and test a wide range of state of the art methods focused in faceswap and facial rejuvenation, establish a train and test pipeline for the selected ones (based on DeepFakesLab) and integrate the solutions in the actual Nuke VX environment. Allowing the graphical designers to use the Deep Learning techniques as internal Nuke modules supported by the company servers infrastructure.

We define a collaborative workflow with two of the most relevant Spanish universities, providing the researchers useful tools to train the deep learning models, as well as, a deployment pipeline to integrate the research in real time surveillance applications.