Instruction

Check

• 'Semantic Instance Segmentation with a Discriminative Loss Function'
• 'ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation'
• 'Towards End-to-End Lane Detection: an Instance Segmentation'

File information

• instance_seg_models_enet_train.py
  Main train model script.

• batch_norm.py
  class for using batch normalization.

• config_etc.py
  for configure some params to train or something.

• DataGen.py
  class for create data batches or load images.

• method.py
  methods like convolution function, etc..

• placeHolders.py
  place holder class for some params. using tf.placeholder(...)

• sementic_seg_models_xxx.py
  tensorflow train architecture for semantic segmentation.
  this models use A1(original images) for segmentation.
  the output(predict_train) images will be saved in _A1_predict_XXX folder.

'xxx' means architecture of models.
 -deeplabv1, enet, etc..

• semantic_seg_apply_crf.py
  crf applied images in A1_predict and save image into A1_predict_crf

Data folder information

• A1
  This folder is original Data set for Instance segmentation.

_centers.png : Center of each leaf.
_fg.png : Sementic segmentation label.
_rgb.png : instance segmentation label.