Skip to content

Latest commit

 

History

History
40 lines (28 loc) · 1.26 KB

README.md

File metadata and controls

40 lines (28 loc) · 1.26 KB
Raw

Hand Signs Classification Competition from DACON

I stand in 39th out of 150 persons.

Competition link is https://dacon.io/competitions/official/235896/overview/description

The Mission was Classifying hand sign images that express 0 to 10 seperately.

image

Key Idea

I used 'Ensemble Train', 'Image augmentaiton'

Actually the number of train images was about 860, so it was necessary to augmentate images.

Frame works

Pytorch for ensemble train.

Albumentations for image augmentation.

Code example

# Load efficientnet_b3 model

loaded_model = torch.load('/content/drive/MyDrive/DACON_Image/weights/b3_model.pt')
model_b3 = Network_b3().to(device)
model_b3.load_state_dict(loaded_model['model_state_dict'])

# Load wide_resnet50_2 model

loaded_model = torch.load('/content/drive/MyDrive/DACON_Image/weights/wrn_model.pt')
model_wrn = Network_wrn().to(device)
model_wrn.load_state_dict(loaded_model['model_state_dict'])

How to use

  1. Download the images from DACON.
  2. Download Ensemble_Train.ipynb.
  3. Take them in to same colab directory.
  4. Activate Ensemble_Train.ipynb and change the directory path to your own path
  5. Run the code