Install git lfs from here to clone the repository.

The solution files are organized in two directories docker and cli. This provides two ways for reproducing our final results for kaggle.

NOTE: Using docker is our recommended way to reproduce our competition results.

1. Unzip the test images stage_2_test_images.zip to a convenient path. <path>/stage_2_test_images will be used in all the examples to refer to test images path.

2. Unzip the train images stage_2_train_images.zip to a convenient path. <path>/stage_2_train_images will be used in all the examples to refer to train images path.

3. Place models from our_weights folder in a convenient path. <path>/weights will be used in all the examples to refer to model/weights path. You can also leave the weights in the current path in the repository.

NOTE: To clone the repository, make sure to have git lfs installed.

• Ubuntu (16.04 recommended)
• 4 GPUs (NVIDIA GeForce GTX 1080Ti or up)
• Compatible nvidia drivers
• docker (version 18.06.1+ recommended)
• nvidia-docker v2.0+
• bash v4.3+
• git lfs

• Ubuntu (16.04 recommended)
• At least 1 GPU (NVIDIA GeForce GTX 1080Ti or up)
• Compatible nvidia drivers
• docker (version 18.06.1+ recommended)
• nvidia-docker v2.0+
• bash v4.3+
• git lfs