What Is Wrong With Scene Text Recognition Model Comparisons? Dataset and Model Analysis (ICCV 2019 Oral)

Jeonghun Baek, Geewook Kim, Junyeop Lee, Sungrae Park, Dongyoon Han, Sangdoo Yun, Seong Joon Oh, Hwalsuk Lee

arXiv Github

Motivations for this Research

RegularIrregular

Examples of regular (IIIT5k, SVT, IC03, IC13) and irregular (IC15, SVTP, CUTE) real-world datasets

Referred to as scene text recognition (STR), reading text in natural scenes, as shown above, has been an essential task in many industrial practices. In recent years, researches have proposed an increasing number of new scene text recognition (STR) models, with each model claimed to have widened the boundary of technology.

While existing methods have pushed the boundary of technology, means for holistic and fair comparison have been, in large, missing in the field because of the inconsistent choices of training and evaluation datasets. It is not easy to determine whether and how much the new module has improved upon the current art, because of the different assessment and testing environments that make it challenging to compare reported numbers at face value.

Problem: Inconsistent Comparison

The table exhibits the performance of existing STR models with their inconsistent training and evaluation settings. This inconsistency hinders the fair comparison among those methods. We present the results reported by the original papers and also show our reimplemented results in a unified and consistent setting. In the last row, we also show the best model we have found, which shows competitive performance to state-of-the-art methods.
The top accuracy for each benchmark is shown in bold.

We examine different training and evaluation datasets used by prior works and point out the discrepancies. We aim to highlight how each work differs in constructing and using their datasets and scrutinize the bias in comparing performance among different works.

Read More

AI RUSH 2019

On June 26, Naver and LINE announced a global hackathon for artificial intelligence, AI RUSH. As of August 29, the two-month-long journey of AI RUSH 2019 has finally come to an end.

Naver and LINE hosted this hackathon to spark AI technology development through a sharing of knowledge and experience in the relevant fields between the participants from all over the world and seasoned developers from Naver and LINE. Current engineers of Naver and Line also participated in this event as mentors, helping the applicants enhance their understanding of on-site knowledge.

Read More

Global Residency Program

2019 Spring NAVER Clova AI Research Global Residency Program

Deep learning and AI technologies have already made their way into our daily lives. They understand our language; recommend music; suggest and summarize news; tell us about a random flower we photographed; translate menus and signs in foreign languages; enable us to read documents and publications in other languages with much greater ease. These technologies will continue to permeate further and deeper into our lives, transforming all aspects of our world. With the dawn of this new era, being able to conduct research and develop these deep learning and AI technologies is essential for anyone looking to truly understand what this exciting new future can bring. However, as challenging as these innovations are, we believe that people with a solid background in basic mathematics and programming can learn these technologies at a high level in just six months to a year. In fact, in that amount of time, it is possible to learn enough to start creating new models and algorithms. It is possible to learn at this level — however, it requires a researcher being willing to commit fully for a minimum of six months.

The NAVER Clova AI Research Residency Program is designed to develop motivated talent, providing a six-month, all-in-one environment for carrying out research and development on deep learning and AI. Participants will be given a research setting in which they can concentrate solely on research into deep learning and AI technologies, backed by a diverse array of real-world data and abundant GPUs, together with guidance from a team of experienced Clova AI Research research engineers and some of the most renowned researchers in the field, including Prof. KyungHyun Cho (NYU), Dr. Jun-Yan Zhu (MIT, 1st author of CycleGAN), and Prof. Hannaneh Hajishirzi (UW). Graduates of this program will be able to develop these cutting-edge technologies that are being applied directly to a whole range of next-generation services, shaping the experiences of hundreds of millions of NAVER and LINE users worldwide. Through the participation of enthusiastic, creative people like you, we hope to usher in the birth of an amazing new era of deep-learning and AI technologies.

Program Overview

  • Duration: January 1, 2019 – June 30, 2019 (6 months) *exact dates negotiable
  • Location: Clova AI Research Team, NAVER Green Factory, Seongnam, South Korea
  • Application submission: clair-reg@navercorp.com
  • Deadline: 15 Nov 2018 11:59 p.m. (AOE)
  • * Interviews will be arranged on a first-come-first-served basis
  • Interview schedule: From 15 Sep 2018 (interview process may vary depending on schedule and circumstances)

What the Clova AI Research Residency Offers

  • A six-month contract position as a member of the Clova AI Research Team, based at NAVER Green Factory
  • Mentoring by Clova AI Research team members and renowned researchers, including Prof. Kyung Hyun Cho (NYU), Dr. Jun-Yan Zhu (MIT, author of CycleGAN), and Prof. Hannaneh Hajishirzi (UW)
  • Competitive salary (equivalent to that of a full-time NAVER employee)
  • Access to NAVER data and research
  • Access to GPU cloud resources via platforms like NSML (refer to https://deview.kr/2017/schedule/186)

Eligibility

  • Bachelor’s degree or higher in Computer Science, Electronics, or Mathematics (people expecting to graduate before 2020 are also eligible)
  • Basic knowledge of and programming ability in probability, information theory, and linear algebra
  • Ability to understand and implement practice problems from Deep Learning Lessons
  • (see https://hunkim.github.io/ml/http://bit.ly/PyTorchZeroAll)
  • Ability to implement recent publications (2016 ~ 2018) related to deep learning/AI
  • English communication skills (reading, writing, speaking)
  • Experience in research/development activities, and publication of papers related to machine learning
  • Awards related to programing and/or mathematics

How to Apply (in English): submit a brief resume along with the following information, all as a single PDF file

  • Introduction to projects you have participated in, and your github information
  • Names and contacts of two or more referees (academic advisor, the supervisor from the previous workplace, etc.) who can write a recommendation letter
  • Description of why you should be selected for the program (maximum one page)
  • Outline of the project that you wish to carry out (maximum one page, focusing on problem definition and data description)
  • Other experiences and achievements that demonstrate your strengths

Internship Program

Introduction

  • Clova AI Research is a global team for research and development of preceding AI technologies for Clova. We are currently looking for AI researchers and engineers with outstanding competencies to make Clova a globally recognized AI platform
  • Clova AI Research has established a very powerful and open collaborative network not only within Clova but also with other organizations within Naver
  • As Clova AI Reseach is a global team, English is the working language

Eligibility

  • Applicants with AI-related majors (machine learning, natural language processing, computer vision, applied mathematics, etc.) or equivalent experience/competency
  • Development capabilities based on the open-source framework, such as Tensorflow, PyTorch, MXNet, Caffe2, etc.
  • Ability to actively identify and conduct self-directed researches

Preferential qualifications

  • English communication skills
  • Ability to quickly and accurately implement the latest works on AI (within 1~2 weeks)

People we want to hire

  • Applicants who enthusiastically perform tasks and missions in a self-directed manner
  • Applicants who can actively communicate and cooperate with an open-mind
  • Nationality is irrelevant
Selection process and other matters
  • Document screening > Coding test > Deep learning implementation test and oral presentation >  Technical interview > Admission (research intern only)
  • Mandatory documents to submit: CV, Research proposal (mandatory for research intern only; for development intern, research proposal is not mandatory)
  • The selection process may vary by schedule and circumstances.
  • A detailed schedule for the selection process will be notified to each applicant, upon finalization.
Additional Information
  • The final selection of an applicant may be canceled if the application he/she submitted is found to contain false information.
  • Those who are eligible for employment protection (veterans, persons with disabilities) will be subjected to preferential treatment according to the relevant laws and regulations.
  • If you have any questions, please send us an email to clova-jobs@navercorp.com.
  • This recruitment may be terminated prematurely once the positions are filled.
  • Please send your CV and cover letter to clova-jobs@navercorp.com.

Full-Time Positions

Introduction

  • Clova AI Research is a global team for research and development of preceding AI technologies for Clova.
    We are currently looking for AI researchers and engineers with outstanding competencies to make Clova a globally recognized
    AI platform.
  • Clova AI Research has established a very powerful and open collaborative network not only within Clova but also with other organizations within Naver.
  • As Clova AI Reserach is a global team, English is the working language.

Eligibility

  • Applicants capable of carrying out preceding researches on AI technology, to improve Clova performance
    (All fields of machine learning-based AI technology including natural language processing, computer vision, and recommendation)Research Scientist
    • Understanding of the latest AI research technology, excellent research capabilities and performance
    • Strong R&D leadership in managing projects and suggest research directionsAI SW Engineer
    • Development capabilities based on the open-source framework, such as Tensorflow, PyTorch, MXNet, Caffe2, etc.
    • Experience in multi-GPU and high-performance computing
    • Experience in designing and constructing machine learning/deep learning models
    • Ability to quickly and accurately implement the latest works on AI

Preferential qualifications

  • Research Scientist
    • Ph.D. degree in AI-related fields (machine learning, natural language processing, computer vision, applied mathematics, etc.) or equivalent experience/competency
    • Publications in top-tier academic associations/conferences in the field of AI
      (NIPS, ICML, ICLR, CVPR, ACL, EMNLP, ICCV, AAAI, IJCAI, KDD, etc.)
    • Excellent English writing and communication skillsAI SW Engineer
    • Experience in CUDA-based GPU programming or distributed processing programming such as Hadoop or Spark
    • Experience in visualization and front-end development

People we want to hire

  • Applicants who enthusiastically perform tasks and missions in a self-directed manner
  • Applicants who can communicate actively and cooperate with an open-mind
  • Nationality is irrelevant
Selection process and other matters
  • Document screening > First interview > Second interview > Admission
  • The selection process may vary by schedule and circumstances.
  • The schedule will be announced to applicants as soon as they are confirmed.
Additional Information
  • The final selection of an applicant may be canceled if the application he/she submitted is found to contain false information.
  • Those who are eligible for employment protection (veterans, persons with disabilities) will be subjected to preferential treatment according to the relevant laws and regulations.
  • If you have any questions, please send us an email to clova-jobs@navercorp.com.
  • This recruitment may be terminated prematurely once the positions are filled.
  • Please send your CV and cover letter to clova-jobs@navercorp.com.