The deep learning artificial intelligence revolution is here and one area that has seen startling advancement is deep learning powered computer vision and image analytics. The advent of transfer learning has made it easier than ever to leverage these advances with limited amounts of labeled data. Despite this, actually integrating deep learning into a workflow is still challenging without deep subject matter expertise.
Cortex is a robust image analytics platform that empowers users to build custom image analytics pipelines from data ingestion to model deployment. Cortex is powered by ImageVec, Vyasa’s proprietary pipeline of image preprocessing and deep learning algorithms; upload raw image data, preprocess, annotate it, train custom models for tasks such as image classification, object detection and segmentation, and more. Designed with collaboration in mind, Cortex’s model management system makes training, analyzing, documenting, and deploying models intuitive. Visualize the results of analysis and then share your work with your team using the built in chat and collaborative analytics module.
Leveraging the power of deep learning for image analytics and computer vision has never been more intuitive.
Features & Benefits
Cortex was built by data people, for data people. It combines a robust model management system with an intuitive user interface that enables customization of preprocessing steps and model hyper-parameters to deliver a powerful modeling platform. Cortex uses Layar, our enterprise data fabric solution, to securely ingest and store user data. Once the data is in Layar it can be easily annotated, preprocessed, and used to train a model in Cortex.
Cortex is powered by ImageVec, a proprietary suite of preprocessing pipelines and deep learning algorithms that handles everything from preprocessing raw image data to delivering a deployable model.
Our custom-built preprocessing pipeline includes a wide range of options, including:
- Full image and localized tiling
- Text scrubbing and parsing
- Various sampling techniques
- A range of data augmentation techniques
- Image vectorization
Cortex provides a robust user interface for the annotation of images by multiple users. Additionally, it supports annotation tracking across users and filtering images by annotations. Once the data has been preprocessed and annotated, users can choose from a wide variety of prebuilt deep learning algorithms. Cortex leverages transfer learning to bring high level performance to problems, even when starting with a small amount of annotated data.
Fine tune a state-of-the-art, pre-trained model for downstream tasks, including:
- Image reconstruction
- Image classification, with or without localization
- Object detection
- Image segmentation
Cortex supports training multiple models in parallel across multiple GPU instances, and offers a range of hyper-parameter optimization capabilities, including bayesian optimization. Watch every epoch in real time with simple yet informative visualizations. Tuning models has never been easier and more efficient!
Training and tuning models is only the first step in leveraging data science, and Cortex doesn’t stop there. Manage, analyze, and store models, along with model meta-data, and deploy your best models with Cortex’s model management system.
Each model is monitored closely throughout training and then deployment. Quickly access who trained the model, when it was trained, the data it was trained on, the preprocessing steps that were taken, the images it’s made predictions on, and the outcomes.
Take model management a step further and combine trained models into an ensemble that work together to make predictions.
The key to data driven communication is visualization, and Cortex allows users to visualize every step in the modeling process. Visually monitor training and then look through the results, sharing what you find with teammates.
Good data science strategy demands reproducibility and documentation, and Cortex gives users the ability to visualize the history of experiments performed over the course of a project.
Easily export images to create your own story, with all the visual evidence already at your fingertips.
Data science is a team sport, and Cortex was designed from the beginning as a collaborative platform.
- Share models, model statistics, and visualizations.
- Built in chat module allows direct collaboration with team members.
- Track activity by user and follow experiments from start to finish.
- Robust account management ensures that users receive access to the features and data they require.
Detection of breast cancer on screening mammography is challenging as an image classification task because cancerous tissue only represents a small portion of the tissue in the image. Cortex rises to the challenge with localized tiling to deliver state of the art results.