In an era of fragmented reading, fewer and fewer people pay attention to the exploration and thinking behind each paper.

In this column, you will quickly get the highlights and pain points of each selected paper and keep up with the cutting-edge AI achievements.

Click “Read the original article” at the bottom of this article to join the community now and see more of the latest paper recommendations.

55

InsightFace

# Open source face recognition library based on MXNet

InsightFace is the open source implementation of DeepInsight Lab’s paper ArcFace: Additive Angular Margin Loss for Deep Face Recognition. This work improved MegaFace accuracy to 98%, surpassing the 91% record held by Vocord of Russia.

In addition, the project includes packaged and aligned face training data (MS1M and VGG2), network frameworks (ResNet, InceptionResNet_v2, DenseNet, DPN and MobiletNet), and loss design (Softmax, SphereFace, AMSoftmax, CosFace and Triplet Loss).

From there, researchers can focus on experimenting with algorithms for face recognition, and industry can easily train them according to their own needs or productize them with high-precision pre-training models provided by the project.

The paper links

https://www.paperweekly.site/papers/1785

Project link

https://github.com/deepinsight/insightface

#Python image enhancement library

Augmentor is a Python image enhancement library, which reduces the complex process of using the image library to write your own code, and can batch complete the image rotation, zoom, zoom, add noise to expand the amount of data.

▲ Input image

▲ Output image

Project link

https://github.com/mdbloice/Augmentor

# Sequence distance measurement

TextDistance is a Python library of 30+ algorithms for calculating distances between two or more sequences.

The project features are as follows:

  • 30 + algorithm

  • Pure Python implementation

  • Easy to operate

  • Supports two or more sequence comparisons

  • Some algorithms have multiple implementations

  • Support Numpy for maximum speed optimization

Project link

https://github.com/orsinium/textdistance

Neural Network Voices

# Talk like Kate Winslet

This project is the corresponding code of the neural network speech synthesis teaching video published by Siraj Raval on YouTube. How to use the deep neural network to transform the voice of ordinary people into the voice of Kate Winslet, a famous British actress. The data set used in this project is an audio book read aloud by Kate Winslet.

The teaching video

Project link

https://github.com/llSourcell/Neural_Network_Voices

Personae

# Application of reinforcement and supervision learning in financial markets

Personae implements deep reinforcement learning, supervised learning algorithms and papers based on TensorFlow and PyTorch, and attempts to apply them to financial markets (stock markets). The algorithms implemented in this project include DDPG, Policy Gradient and DualAttnRNN.

▲ Comparison of stock trading returns

▲ Stock price forecast results

Project link

https://github.com/ceruleanacg/Personae

NNDial

# Open source toolkit for end-to-end conversation systems

NNDial is an open source toolkit for building an end-to-end trainable task-based dialogue model. This project is from the University of Cambridge, using a dataset called CamRest676.

▲ Test results

Project link

https://github.com/shawnwun/NNDIAL

Voice Activity Detection Toolkit

# Voice endpoint detection kit

This project is an open source implementation of the paper Voice Activity Detection Using an Adaptive Context Attention Model, and also includes a speech data set recorded by the author team.

The toolkit supports four MRCG-based classifiers:

  • Adaptive Context Attention Model (ACAM)

  • Enhanced Deep Neural Network (bDNN)

  • Deep Neural Network (DNN)

  • Cyclic Neural Network Based on LSTM (LSTM-RNN)

Project link

https://github.com/jtkim-kaist/VAD

Knowledge Graph Representation

PyTorch implements knowledge graph representation

The project organized four commonly used data sets for knowledge graph representation, provided code for data cleaning and collation, and implemented four translation based algorithms with PyTorch. During the evaluation phase, multi-process acceleration was used to compress the evaluation time of MeanRank and Hits@10 to less than 1 minute.

Project link

https://github.com/jimmywangheng/knowledge_representation_pytorch

PyHanLP

#HanLP Python interface

This project is a Python interface for The Chinese language processing package HanLP, supporting automatic download and upgrade of HanLP, compatible with PY2 and PY3.

HanLP is a Java toolkit consisting of a series of models and algorithms that aims to popularize natural language processing in production environments. HanLP has the features of perfect function, high performance, clear architecture, up-to-date corpus and customizable.

HanLP can provide Chinese word segmentation, partof speech tagging, named entity recognition, keyword extraction, text recommendation, dependency parsing, text classification, WORD2VEc and corpus tools.

HanLP home page

HanLP自然语言处理包开源

Project link

https://github.com/hankcs/pyhanlp

This paper is selected and recommended by PaperWeekly, an AI academic community, which covers natural language processing, computer vision, artificial intelligence, machine learning, data mining and information retrieval. Click “Read the original article” and join the community immediately!

#Creators recruit #

Let your words be seen by many, many people, like us as join us

I am a egg

Unlock new features: Popular job recommendations!

PaperWeekly applets updated

Today’s arXiv√ Guess you like √ Hot jobs √

Getting a full-time job or an internship isn’t a problem

 

Unlock the way

1. Identify the qr code below to open the mini program

2. Log in with PaperWeekly community account

3. You can unlock all functions after login

job

Please add small assistant wechat (PWBOT02) for consultation

 

Long press to identify the QR code, using the mini program

* Click to read the original article to register

About PaperWeekly

PaperWeekly is an academic platform for recommending, interpreting, discussing and reporting the cutting-edge papers of artificial intelligence. If you research or engage in the FIELD of AI, welcome to click “Communication Group” in the background of the official account, and the little assistant will bring you into PaperWeekly’s communication group.

Del | click read | join community brush in the original paper