rctzeng / GAIN

implementation of Guided Attention Inference Network

GAIN-tensorflow [under construction]

Note: the code is not stable and some functions are still under construction

The repository holds the implementation of GAIN, and the segmentation network mainly borrows the code from from xtudbxk/SEC-tensorflow.

Prerequisite

• Tensorflow >= 1.5
• Pydensecrf >= 1.0
• Download the Pascal VOC data and the pre-train VGG16 model, please refer to step 2 in xtudbxk/SEC-tensorflow

Train GAIN network

• Training: python [model].py -g <gpu_id> -f <gpu_fraction> <-c if you'd like to train with CRF>
• Predicting Mask: python [model].py -g <gpu_id> -f <gpu_fraction> -a inference