/NAVER_Boostcamp_OCR_competition

p4-fr-ocr-oriental-chicken-curry created by GitHub Classroom

Primary LanguagePython

Pstage 4 ] OCR

πŸ“‹ Table of content



πŸ‘‹ νŒ€ μ†Œκ°œ

  • OCR 7μ‘° oriental-chicken-curry
  • 쑰원 : κΉ€μ§„ν˜„, 김홍엽, κΉ€νš¨μ§„, 김희섭, λ°•μ„±λ°°
κΉ€μ§„ν˜„ 김홍엽 κΉ€νš¨μ§„ 김희섭 λ°•μ„±λ°°



πŸŽ– μ΅œμ’… κ²°κ³Ό

  • Ranking : 2/12
  • Score
    • Public : 0.8170
    • Private : 0.6065



λŒ€νšŒ κ°œμš”



μˆ˜μ‹ 이미지λ₯Ό latex 포멧의 text둜 λ³€ν™˜ν•˜λŠ” λ¬Έμ œμž…λ‹ˆλ‹€. μˆ˜μ‹ μΈμ‹μ˜ κ²½μš°λŠ”, 기쑴의 κ΄‘ν•™ 문자 μΈμ‹κ³ΌλŠ” 달리 multi line recogintion을 ν•„μš”λ‘œ ν•©λ‹ˆλ‹€.

κΈ°μ‘΄ single line recognition 기반의 OCR이 μ•„λ‹Œ multi line recognition을 μ΄μš©ν•˜λŠ” κΈ°μ‘΄ OCRκ³ΌλŠ” μ°¨λ³„ν™”λ˜λŠ” Taskμž…λ‹ˆλ‹€.

  • 평가방법
    • sentence accuracy, wer
    • score : sentence accuracy * 0.9 + wer * 0.1



πŸ“ 문제 μ •μ˜ 및 ν•΄κ²° 방법

  • ν•΄λ‹Ή λŒ€νšŒμ— λŒ€ν•œ 문제 μ •μ˜, ν•΄κ²° 방법, μ›Ή μ„œλΉ™ λ“±μ˜ λ‚΄μš©μ€ μ—¬κΈ°μ„œ μžμ„Έν•˜κ²Œ 확인 ν•΄ 보싀 수 μžˆμŠ΅λ‹ˆλ‹€.

  • ν˜‘μ—… κ΄€λ ¨ λ‚΄μš©μ€ μ—¬κΈ°μ„œ 확인 ν•  수 μžˆμŠ΅λ‹ˆλ‹€

πŸ’» CODE μ„€λͺ…

β”œβ”€β”€ README.md
β”œβ”€β”€ configs           # yaml -> νŒŒλΌλ―Έν„° μˆ˜μ • 
β”œβ”€β”€ data
β”œβ”€β”€ data_tools
β”œβ”€β”€ inference&practice
β”œβ”€β”€ inference.py      # λͺ¨λΈ μΆ”λ‘ 
β”œβ”€β”€ log.py
β”œβ”€β”€ networks          # SATRN , SRN λ“± OCR λͺ¨λΈ
β”œβ”€β”€ requirements.txt  
β”œβ”€β”€ train.py          # ν•™μŠ΅ μ½”λ“œ
β”œβ”€β”€ unit_test.py      # test μ½”λ“œ
└── utils             # κ·Έ μ™Έ μœ ν‹Έ μ½”λ“œ
  • Train & Test code
python code/train.py

python code/inference.py