minar09/U-Net-Attention

U-Net + Attention, extending U-Net model for semantic segmentation. Implemented with TensorFlow.

Fashion parsing models in TensorFlow

1. Tensorflow implementation of Fully Convolutional Networks for Semantic Segmentation (FCNs).
2. TensorFlow implementation of U-Net

The implementation is largely based on the reference code provided by the authors of the paper link.

1. Prerequisites
2. Training
3. Testing
4. Visualizing

Prerequisites

• pydensecrf installation in windows with conda: conda install -c conda-forge pydensecrf. For linux, use pip: pip install pydensecrf.
• Check dataset directory in read_dataset function of corresponding data reading script, for example, for LIP dataset, check paths in read_LIP_data.py and modify as necessary.

Training

• To train model simply execute python FCN.py or python UNet.py
• You can add training flag as well: python FCN.py --mode=train
• debug flag can be set during training to add information regarding activations, gradients, variables etc.

Testing

• To test and evaluate results use flag --mode=test
• After testing and evaluation is complete, final results will be printed in the console, and the corresponding files will be saved in the "logs" directory.

Visualizing

• To visualize results for a random batch of images use flag --mode=visualize