This repository contains the TensorFlow implementation of paper ESPNet.

The code is tested using Tensorflow v1.8 with Python v2.7.

ISIC 2017: Skin Lesion Analysis Towards Melanoma Detection

• dataset.py Source code for preparing dataset. If you are not using the ISIC 2017 skin segmentation dataset, kindly make required changes in the dataset.py file.
• train.py Source code for training ESPNet-C and ESPNet model.
• model.py Source code for model architecture of ESPNet-C and ESPNet model.
• config.py Source code for configuration file containing image size required for training.
• test.py Source code for creating segmentation masks of images for testing.
• eval.py Source code for evaluating model.

The network can be trained using the train.py script.

python train.py --model_file_name <model file name> \
--epochs <number of epochs> \
--batch_size <batch size> \
--model_name <espnet_c/espnet>\
--image_dir <Folder containing training images>\
--ann_dir <Folder containing annotations of training images>

Segmented mask can be generated using test.py script.

python test.py --model_file_name <model file name> \
--batch_size <batch size> \
--input_folder <Folder containing images to be tested> \
--op_folder <Output Folder> 

Model can be evaluated using eval.py script.

python eval.py --ground_truth <ground truth directory> \
--prediction <prediction image directory containing images of mask predicted by the model>