Cleaner data and better models.
Today, without the code.
Build models with platform today or use a competition to get it done quickly
Platform is a human-in-the-loop development environment to train computer vision models. Models are trained via interactions between the human user and the visual environment, which employs advanced deep learning to create unique visual representations of data provided to it.
How does it work?
With our proprietary technology, we help domain experts identify patterns in the data by displaying visual groupings or clusters that can easily be labelled. These labels are used by a backend model that can be trained and fine-tuned in minutes. The trained model alters the degree of separation between visual clusters and the process repeats itself to allow an iterative style of model development.
Upload
Upload data onto our servers. Uploading labels for data at this stage is completely optional since our iterative process allows users to label their data iteratively. For large datasets, we offer direct linking between servers.
Examine
Our visual representation environment allows users to switch between ‘views’ of data. The views are 3D projections of higher-dimensional data mapped onto a blank canvas. The regions in the ‘canvas’ are semantically related with visually similar objects being closer together. In the backend, we selectively combine data from hundreds of dimensions and then exhibit them in 3d space for easy visual segregation.
Label
Labelling data on platform is incredibly easy with the label-as-you-go strategy. All it takes is drag-selecting images on the visual environment or ‘canvas’ and then creating a new label or selecting an old one. Combined with incremental training, this method yields the best models in the shortest possible time frames.
Train
Once sufficient data is labelled, a user can train the backend model further. The results of training are immediately interpretable in the form of changes in separation between visual clusters. Our empirical rule is that greater cluster separation yields better models.
Deploy
Backend models can be downloaded for downstream tasks with the click of a button. That isn’t the end though; with platform, you can just as easily download labels for all your data in a single click. These labels are predicted via the background model with accuracy that generally exceeds human performance.
Point and click tools
Platform tools allow AI model development with point and click actions.
Platform tools allow user to quickly work through all the phases of AI model through simple point and click actions. With platform, domain experts can quickly upload data to obtain cleaner labels and AI models that can be passed on to production.
Accelerated Development
This template uses such features as Global swatches and Symbols. It gives you the ability to customize the template for your need easily and fast.
With Platform, AI model development speeds up by several orders of magnitude. The main advantage of platform is that it offers users a unified workflow environment – one that allows users to work on data and models simultaneously.