ALASKA2 Image Steganalysis

Source code of solution for ALASKA2 Image Steganalysis competition.

Solution

Key points:

• Efficientnets
• DDP training with SyncBN and Apex mixed precision
• AdamW with cosine annealing
• EMA Model
• Bitmix

Quick setup and start

Requirements

• Nvidia drivers, CUDA >= 10.2, cuDNN >= 7
• Docker, nvidia-docker

The provided dockerfile is supplied to build image with cuda support and cudnn.

Preparations

• Clone the repo, build docker image.

  git clone https://github.com/lRomul/argus-alaska.git
  cd argus-alaska
  make build
  

• Download and extract dataset to data folder.

Run

• Run docker container

make

• Create folds split and extract quality of images

python make_folds.py
python make_quality_json.py

• Train model

python train.py --experiment train_001

• Predict test and make submission

python predict.py --experiment train_001

• Train on 4 GPUs with distributed data parallel

./distributed_train.sh 4 --experiment ddp_train_001