kaggle_rsna

Requirements

• Python 3
• https://github.com/tensorflow/models

How to

Solution is under heavy development and baseline not working yet. Use at your own risk and expect waste of time.

1. Preprocess data

Test data

python convert_data_tf_format.py --input_images_path rsna_data/stage_1_test_images --input_labeling_path rsna_data/stage_1_test_images.csv --output_path rsna_data_preprocessed/stage_1_test_images.1.tfrecord --threads 7

Train data

python convert_data_tf_format.py --input_images_path rsna_data/stage_1_train_images --input_labeling_path rsna_data/stage_1_train_labels.csv --output_path rsna_data_preprocessed/stage_1_train_images.tfrecord --threads 7

2. Make train/eval split

python split_train_eval.py --input_tf_record rsna_data_preprocessed/stage_1_train_images.tfrecord --input_labeling_path rsna_data/stage_1_train_labels.csv --input_detailed_info rsna_data/stage_1_detailed_class_info.csv --output_prefix rsna_data_preprocessed/stage_1_train_images

3. Train baseline

Run as follows:

python ${TENSORFLOW_MODELS_REPO}/research/object_detection/model_main.py --alsologtostderr --pipeline_config_path models/baseline_model/ssd_mobilenet_v1_focal_loss.config --model_dir models/baseline_model

where

• ${TENSORFLOW_MODELS_REPO} - path to your copy of https://github.com/tensorflow/models