kaggle-human-protein-atlas-image-classification

Kaggle 2018 @ Human Protein Atlas Image Classification

35th Place Algorithm and Model.

'*' : Working methods between options.

Models

• vgg16
• resnet50, resnet101, ...
• densenet121, densenet169 *
• inception v3, inception v4 *
• se152
• polynet
• NASNet, PNASNet

Implementations

• Data Loader for External Datas and Merger
• Basic data augmentations
  • Rotation, Flip *
  • Channel drops
• 16 Test-Time Augmentation
• 5-Folds Cross Validation
• Simple Threshold Search Algorithm
• Ensembles
  • Test-Time Augmentation Averaging *
  • Majority Voting *
  • Fully-Connected Neural Network
    • logits -> output
    • logits + features -> output
  • XGBoost
• Loss
  • Soft F1 Loss *
  • Binary Cross Entropy *
  • Focal Loss
  • MultiLabelMarginLoss

Train

Training with yaml configuratin files and changed few parameter.

$ python main.py -c conf/densenet.yaml
$ python main.py -c conf/densenet.yaml --lr 0.0001