Mccn trainer

• Adds num_worker option to trainer
• Adds a logger to save losses, images, and gradients which can be visualized with tensorboard
• Adds a new architecture of mccnn with shared layers, MCCNNv2 and its extractor
• Added a RandomHorizontalFlipImage class for Multi channel images which works with numpy.ndarrays (solves the requirement to transform to PIL image for using the torchvision version, which doesnt work for num_channels>4)

bob/learn/pytorch/datasets/__init__.py

@@ -11,6 +11,7 @@ from .utils import Normalize
 from .utils import Resize 
 from .utils import ToGray
 from .utils import ChannelSelect 
+from .utils import RandomHorizontalFlipImage
 
 from .utils import map_labels
 from .utils import ConcatDataset