Fishchecker

2022 데이터 청년 캠퍼스 고려대학교 빅데이터 기반의 지능정보시스템 개발 : Team 11

• Fishchecker
  • Development Environment
  • How to use
    • Colab
    • PyQt5 UI
  • Results for some Test cases
    • Colab test version
    • PyQt5 UI
  • EDA
  • Algorithm
    • Future Direction of Algorithm Time Complexity Improvement
  • Team members
  • 도움주신 분들

The problems to be solved in this project are as follows. Fish with the camera function of the mobile when the picture was taken,

1️⃣ Classify the fish in the picture
2️⃣ Predict the weight of a fish based on an existing data set
3️⃣ Provide meaningful information to users.

This project provides meaningful information to users based on fish species classification and weight prediction functions. Based on the mobile application service, users can conveniently Its purpose is to make it possible to obtain information.

Development Environment

• EDA : R Studio
• System : VSCode, Colab PRO

How to use

Colab

NOTE: PyQt5 UI는 사용환경에 따라 실행시간에 차이가 클 수 있습니다. 또한 영상실행 단계에서 코덱 issue가 발생할 수 있어 fishChcker algorithm 결과를 어떤 사용환경에서도 확인해보실 수 있도록 Colab test version를 지원하고 있습니다.

1. Download the ZIP file from the link below

   https://drive.google.com/file/d/1ixf9Td-opkCBDDwDRdwEQicY2Ljvy700/view?usp=sharing

2. Unzip the downloaded file and upload it to My Drive in Google Drive.

3. For faster execution, change the runtime type to GPU and run all shells

PyQt5 UI

1. Install the codec from the link below for smooth execution in the video execution stage of the UI

   https://codecguide.com/download_kl.htm

2. Download the EXE file and run it

   NOTE: Please make sure that all paths do not contain Korean characters.

3. If you want to directly create and run the executable file or run main.py directly from the shell, please match the versions of the packages you need.

python -m pip install -r requirements.txt

Results for some Test cases

Colab test version

Colab.1.mp4

PyQt5 UI

PyQt5.1.1.mp4

EDA

NOTE: EDA에 대한 detail은 https://github.com/SonHyegang/KData2022_Fishchecker/tree/main/EDA 에서 확인하실 수 있습니다

Algorithm

NOTE: Algorithm에 대한 detail은 https://github.com/SonHyegang/KData2022_Fishchecker/tree/main/Fishchecker%20Colab 에서 확인하실 수 있습니다

Future Direction of Algorithm Time Complexity Improvement

• It was intended to intuitively express that the process of predicting weight through the brute force algorithm is realistically possible > If we add the step of narrowing the range of rotation angle by looking at the aspect ratio of the bounding box, there is no need to look at all -90 to 90 degrees.
• The model load is carried out 3 times, but this can be reduced to 2 (Because the model used in the process of selecting the bestframe and the process of finding the length of the fish in the bestframe is the same)
• Since the length of the video used for the demo is 5 seconds, we calculated quite a lot of frames. Considering the cycle of flopping of live fish, it is judged that the weight can be predicted sufficiently even with an image for 1 to 2 seconds.
• Process after selecting the bestframe (finding the length of the fish in the bestframe, finding the thickness of the handle in the bestframe) parallel processing
• Reparameterization (Before reparameterization, the model has more parameters and computation cost.)

Team members

Member	Contributions	Contact
김지나
손혜강		[email protected]
윤세정
임정아
황영석

최종 시연을 위한 v16_best.pt, bong_v5_best.pt file외 중간산출물과 raw data들은 장시간 data 수집과 training, tuning의 결과물이란 점, 용량제한 문제점으로 인해 Github repository에서 공개하고 있지 않습니다. 필요하신 분들은 Team members table의 contact mail로 연락주시기 바랍니다.

도움주신 분들

한국수산회 수산정책연구소 소장님

해양수산부 운영지원과 관계자

노량진수산물도매시장 경영기획부 관계자