At Ximilar, we create, and continuously improve, advanced visual search & image recognition services for businesses globally. We are also developing custom solutions for healthcare, Industry 4.0, ecommerce, fashion and real estate. We provide several e-commerce features for fashion & many new classification and detection features such as automated image labelling, visual similarity search on generic and product images, fashion search, and more. Our ML/AI models can handle hundreds of labels & predict several images per second.
Ximilar Image Recognition :
Ximilar Custom Image Recognition service (previously Vize.ai) provides a trainable image recognition API to recognize and classify images. It allows you to implement state-of-the-art artificial intelligence (AI) into your project. We provide an simple and intuitive user interface for setting up your projects managing them, uploading images, train new models and evaluating the result. After an easy setup, you get results over API and you are ready to build this functionality in your application. It’s easy, quick and highly scalable.
Speeding up prediction times of complex deep learning models
Our challenge was to create accurate, robust, transparent, and fast deep learning models that can be used across e-commerce, medtech and Industry 4.0. We develop a lot of features that users can use as image augmentation settings, explainability, evaluation of independent datasets, and more , thereby making e-commerce platforms more searchable and intuitive.
We were able to convert image recognition models with Intel OpenVINO toolkit and speed up the inference time of API endpoints for our customers. We compared Intel OpenVINO technology with other similar solutions and the Intel OpenVINO optimized models were noticeably faster.
Firstly, our Fashion Search & Tagging system contains more than 100 models. The models are connected in a service called Ximilar Flow, which can be used to build a complex hierarchical classification system. By using OpenVINO technology we were able to speed up the inference time significantly when compared to our non-optimized Tensorflow models running on 2nd gen Intel Xeon Scalable processors. We are using Intel OpenVINO toolkit and OpenCV inside docker images that are run in our Kubernetes-based cloud. All of our servers are based on Intel Xeon processors.
One of our other customers from MedTech is also using the system for the categorization of filament types from the microscope slide. The company is using our technology at a large scale and with Intel AI, we were able to deploy their models on standard CPUs rather than using dedicated accelerators, thus saving a lot of computing resources that can be used for training.
We are continuing to work together with Intel engineers on our Object Detection, RemoveBG, Segmentation models to make them faster without using GPU.
Our aim is to build the best & easy-to-use platform focused on computer vision, image recognition, detection and search. We are looking forward to converting more of our services and models with Intel OpenVINO tollkit and speed up the entire platform running on Intel Xeon Scalable processors.