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.
I used 'Ensemble Train', 'Image augmentaiton'
Actually the number of train images was about 860, so it was necessary to augmentate images.
Pytorch for ensemble train.
Albumentations for image augmentation.
# 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'])
- Download the images from DACON.
- Download Ensemble_Train.ipynb.
- Take them in to same colab directory.
- Activate Ensemble_Train.ipynb and change the directory path to your own path
- Run the code