Skip to content

Latest commit

 

History

History
39 lines (25 loc) · 1.55 KB

README.md

File metadata and controls

39 lines (25 loc) · 1.55 KB

C23-PC725 Capstone Project

Members

  • (MD) A346DKX4282 - Fernandico Geovardo
  • (ML) M049DSX0588 - Dhuha Ardha Saputra
  • (ML) M181DSX3641 - Vincent Yovian
  • (CC) C368DSX2892 - I Ketut Teguh Wibawa Lessmana P. T.
  • (CC) C220DSY0626 - Audy Revi Nugraha
  • (CC) C303DKY3970 - Vanessa Evlin

Machine learning model

We create a deep neural network for classifying seven types of injuries that often happen in accidents.

Dataset info

We collected our Dataset various wound type from kaggle. We got 431 wound type image data in total and there is the distribution.

  • Abrasions: 85
  • Bruises: 122
  • Burns: 59
  • Cuts: 50
  • Ingrown Nails: 31
  • Lacerations: 61
  • Stab Wounds: 23

Architecture

We use the technique of transfer learning by adding multiple layers of fully connected networks on top of an InceptionV3 model pre-trained on the imagenet dataset. We freeze all the layers in the pre-trained InceptionV3 except for the last 12. We saved our best performing model in .h5 format and its weights in a .ckpt file, which so far has reached a validation accuracy of over 0.83 for the injury classification task and over 0.89 for the accident classification task. The dataset, latest checkpoints, and various model formats can be found here. Legacy versions of the model can be found here.

Cloud computing and deployment

https://github.com/vanessaevlin/tolong-capstone

Mobile app

https://github.com/fernandicogeo/mobile-tolong