Deep Learning Fast and Easy

Why FastEstimator?

FastEstimator is a high-level open source deep learning framework that can help you build complex deep learning systems easily. It provides all the simplicity and flexibility needed to build a high-performance deep learning model and run it anywhere.

Hybrid

Built upon TensorFlow2 and PyTorch, FastEstimator combines the best of the two. Enjoy both speed and flexibility!

Simple...

You only need to know 3 modular APIs to get started. Use your favorite backend for customization, no extra learning is needed.

Yet Flexible!

We introduce new concepts for AI modularization: Operator & Trace, which can help you implement your craziest ideas without friction.

Easy to Scale

Multi-GPU training requires no efforts on your side. Just run your code on a multi-GPU system and we will scale it for you!

Pre-Bundled Power

FastEstimator provides ready-made modular AI components to make prototyping easier. Plug them into your own applications!

Application Hub, not Model Zoo

We offer end-to-end state-of-the-art workflows across different AI domains. Come learn with us and use them for your own projects!

How does FastEstimator work?

All deep learning training workflows involve three essential components: data pipeline, network, and optimization strategy. Each one becomes a fundamental API in FastEstimator:

Pipeline

Pipeline takes care of loading and preprocessing data

Network

Network compiles all trainable and differentiable model operations

Estimator

Estimator manages the training loop

Our Application Hub

FastEstimator does not merely provide a model zoo, but rather end-to-end implementations of state-of-the-art deel learning solutions. Every template has step-by-step instructions to ensure that you can easily build new AI applications using your own data.

Handwritten Digit Classification LeNet, MNIST dataset
Train a LeNet neural network to classify MNIST handwritten digital number.
Lung X-ray Image Segmentation Unet, Montgomery dataset
Train a neural network to label the region of patient lung from a X-ray images.
Fast Style Transfer Modified ResNet, COCO2014 dataset
Train a neural network that applies art style on any target image