weights about smoke-semantic-segmentation HOT 2 CLOSED

@zk2ly The seg_model_weights.best.hdf5 was for model which had the following structure:

• The X in layers.Conv2D(**X**, (3, 3), activation='relu', padding='same') increased as:
  • 8 -> 16 -> 32 -> 64 -> 128 -> 64 -> 32 -> 16 -> 8 [as it should happen in encoder-decoder]
  • As a fix I will update the seg_model_weights.best.hdf5 weights. These may not be the best since I am limited by input dataset and computational resources and it can be definitely improved

from smoke-semantic-segmentation.

A visualization of previous weights (24 layers) by netron

from smoke-semantic-segmentation.