Make triplet loss classifier

[dajinstory]

• standard cnn model with specific loss function(triplet loss)
• object detection model.
• end to end model or 2 models. tba

[dajinstory]

20191212(목)

• 전체 서비스 구상도 픽스

  • server_for_service ubuntu18.04. aws ec2. no gpu.
  • server_for_DL laptops(tf with cpu), dajin's desktop(tf with gpu), AI lab server(tf with gpu)
  • DeepLearning library tensorflow + keras. tensorflow version is 1.14
  • Web Framework Flask + python

• image labeling 툴 환경설정 및 테스트

  • labelImg window + conda + https://github.com/tzutalin/labelImg

• DeepLearning models

  1. First Trial: use two models.
     • get object detection with darkflow
     • get more accurate result with simple CNN model with triplet loss
  2. Second Trial: after the first attempt is successful, try to use R-CNN

[dajinstory]

20191213(금)

triplet loss 코드 작성 yolo training test 완료 data 구성만 하면 바로 트레이닝 가능 모델 병합은 하지 않음