Login With Github

Take Away! The 12 Latest AI Open Source Projects

1. Dopamine

# A reinforcement learning framework based on TensorFlow

Dopamine, which is based on TensorFlow, is a research framework for implementing prototypes of reinforcement learning algorithms quickly. It is designed to provide researchers with an easy-to-use lab environment to meet the needs for small, easy-to-access codebases. With the framework, you will feel easy to build experiments to verify your own thoughts during the research process.

The link for the project:


2. TransmogrifAI

#An end-to-end AutoML library for structured data

TransmogrifAI is an AutoML library, which is written in Scala and runs on Spark, and its open source is provided by Salesforce. This project aims to help developers accelerate the productization process through automatic machine learning technology. With just a few lines of code, data cleanup, feature engineering and model selection will be automated, and then a high-performance model will be trained for further exploration and iteration.

The link for the project:


3. OpenNRE

#A toolkit for neural network relationship extraction

OpenNRE is a neural network relationship extraction toolkit, which is based on TensorFlow and the open source is provided by the team of Liu Zhiyuan (from the Department of Computer Science of Tsinghua University). This project divides the relationship extraction into four steps: Embedding, Encoder, Selector, and Classifier.

The link for the project:


4. TensorFlow Model Analysis

#A TensorFlow model analysis open source library

TFMA is an open-source library from Google, which aims to help TensorFlow users to analyze the trained models. Users can use the metrics defined in Trainer to evaluate models of large amounts of data in a distributed manner. The metrics can be calculated on different pieces of data and the results can be visualized in Jupyter Notebooks.

The link for the project:


5. GraphPipe

#A generic deep learning model deployment framework

GraphPipe is a generic deep learning model deployment framework, and the open source is offered by Oracle. It aims to help users simplify the deployment of machine learning models and free them from the implementation of specific framework models. GraphPine provides a across-deep-learning-frameworks model generic API, out-of-the-box deployment solutions, and powerful performance. It supports frameworks such as TensorFlow, PyTorch, MXNet, CNTK, and Caffe2.

The link for the project:


6. ONNX Model Zoo

#A generic deep learning pre-training model set

This project brings together the current best types of deep learning pre-training models. The models are all launched in the ONNX (Open Neural Network Exchange) format by Facebook and Microsoft, allowing models to be migrated between different frameworks. Each model has a corresponding Jupyter Notebook that contains information such as model training, operational reasoning, datasets, and references.

The link for the project:


7. 106-point face calibration algorithm based on deep learning

#A useful open source face calibration algorithm

The useful open source face calibration algorithm includes pre-processing steps for facial beauty, beauty makeup, Crycocelle vivo detection, and face calibration. The Windows project part is based on the traditional SDM algorithm, and by modifying the open source code to streamline part of the test code and to optimize the code structure. The Android code part is based on deep learning, and an efficient network model which is good at robustness and supports multi-face tracking is designed. At present, the deep learning algorithm has achieved good results in the direction of face calibration. The project aims to provide an implementation approach which is relatively simpler and easier to use.

The features of the project include:

  • 106 points, making the face contour description more delicate
  • high accuracy. You can still achieve a good calibration result in backlighting and dark light conditions.
  • small models. The tracking model is around 2 MB, making it ideal for mobile integration.
  • fast speed. Android platform code is on Qualcomm 820 (st), and it only needs 7ms for a single face.
  • increasing multi-face tracking.

The link for the project:


8. MagNet

#PyTorch-based deep learning API

MagNet is an advanced deep learning API based on the PyTorch package, and it's designed to reduce template code for developers and improve the efficiency of deep learning project development.

The link for the project:


9. NLP.js

#A general NLP toolkit based on Node.js

NLP.js is a general-purpose natural language processing toolkit based on Node.js. Currently, it supports various tasks such as word segmentation, stem extraction, sentiment analysis, named entity recognition, text classification, and text generation.

The link for the project:


10. Texar

#A TensorFlow-based text generation toolkit

Texar is a TensorFlow-based text generation toolkit that supports tasks such as machine translation, dialog systems, text summaries, and language models. Texar is designed for researchers and practitioners and is used for rapid prototyping and experimentation.

The link for the project:


11. Evolute

#An easy-to-use evolutionary algorithm framework

Evolute is an easy-to-use evolutionary algorithm framework. It defines the basic structures such as individuals and populations, and implements the common operators such as Selection, Reproduction, Mutation, and Update in evolutionary algorithms.

The link for the project:


12. Task-Oriented Dialogue Dataset Survey

#A task-driven dialog data collection

This project is a task-driven dialog data collection, and it brings together research data sets for multiple classic task-driven dialog systems such as Dialog bAbI, Stanford Dialog, Consonant data, DSTC-2, CamRest676, and DSTC4.

The link for the project:


0 Comment