This is the implementation of the models and code in paper:

A MINIATURIZED SEMANTIC SEGMENTATION METHOD FOR REMOTE SENSING IMAGE

Shou-Yu Chen, Guang-Sheng Chen and Wei-Peng Jing

Email: (Shou-Yu Chen)[email protected]

Software and hardware:

1. programming language: Python 3.6.
2. deep learning framework: Tensorflow 1.6 and Keras 2.0.
3. main hardware: Macbook Pro 16G, Intel Core i7 3.1GHz, NVIDIA 1080 eGPU (8G).

Prepare the dataset:

1. change 'dataset_dir' in config.py to your dataset root path.
2. In _test_utils.py, change 'city_names_needed' in test_crop_dataset()to the city list you want,
   and 'percent' to the percent of data amount in these cities you want to process, then, run this file to obtain
   dataset which model can train on it.

Train the model

1. you can choose one of 'unet' or 'micro_net' as model in __name__ == '__main__' in train.py.
2. run python3 train.py to start training, Tensorboard log and model weight files will be automatically
   stored in the path defined by 'dataset_dir' in config.py.

Results

1. run tensorboard --logdir=log in the path you save the log to start tensorboard.
2. open your browser and enter the 'http://localhost:6006' to observe the results.