WIP: Generic trainer

• Adds a generic trainer class and trainer script
• Moved the model specific stuff to config files; example config added
• uses .to(device) throughout

bob/learn/pytorch/config/generic/extractor_configs/wmca_deep_mspad.py

+
+
+# =============================================================================
+# define instance of the preprocessor:
+
+from bob.pad.face.preprocessor import VideoFaceCropAlignBlockPatch
+
+from bob.pad.face.preprocessor import FaceCropAlign
+
+from bob.bio.video.preprocessor import Wrapper
+
+from bob.bio.video.utils import FrameSelector
+
+from torchvision import transforms
+
+import numpy as np 
+
+from bob.pad.face.database import BatlPadDatabase
+
+from bob.learn.pytorch.datasets import ChannelSelect, RandomHorizontalFlipImage
+
+
+FACE_SIZE = 224 # The size of the resulting face
+RGB_OUTPUT_FLAG = True # RGB output
+USE_FACE_ALIGNMENT = True # use annotations
+MAX_IMAGE_SIZE = None # no limiting here
+FACE_DETECTION_METHOD = 'mtcnn' # use annotations
+MIN_FACE_SIZE = 50 # skip small faces
+ALIGNMENT_TYPE='default'
+
+
+_image_preprocessor = FaceCropAlign(face_size = FACE_SIZE,
+                                   rgb_output_flag = RGB_OUTPUT_FLAG,
+                                   use_face_alignment = USE_FACE_ALIGNMENT,
+                                   alignment_type=ALIGNMENT_TYPE,
+                                   max_image_size = MAX_IMAGE_SIZE,
+                                   face_detection_method = FACE_DETECTION_METHOD,
+                                   min_face_size = MIN_FACE_SIZE)
+
+_frame_selector = FrameSelector(selection_style = "all")
+
+preprocessor = Wrapper(preprocessor = _image_preprocessor,
+                       frame_selector = _frame_selector)
+
+
+
+
+#====================================================================================
+# DeepPixBiS Extractor
+
+
+from bob.learn.pytorch.extractor.image import GenericExtractor
+
+from bob.learn.pytorch.architectures import DeepMSPAD
+
+
+from bob.bio.video.extractor import Wrapper
+
+MODEL_FILE='/idiap/temp/ageorge/Pytorch_WMCA/Generic-DeepMSPAD/model_0_0.pth'
+####################################################################
+
+SCORING_METHOD='pixel_mean'  # 'pixel_mean','binary','combined'
+
+SELECTED_CHANNELS = [0,1,2,3] 
+####################################################################
+_img_transform = transforms.Compose([ChannelSelect(selected_channels = SELECTED_CHANNELS),transforms.ToPILImage(),transforms.Resize((224,224), interpolation=2),transforms.ToTensor()])
+
+
+
+def extractor_function(output,kwargs):
+
+  output_pixel = output.data.numpy().flatten()
+
+  score=np.mean(output_pixel)
+
+  return score
+
+network= DeepMSPAD(pretrained=True, num_channels=4)
+
+
+_image_extracor=GenericExtractor(network=network,extractor_function=extractor_function,transforms=_img_transform, model_file=MODEL_FILE,scoring_method='pixel_mean')
+
+_frame_selector = FrameSelector(selection_style = "all")
+
+extractor = Wrapper(_image_extracor, frame_selector = _frame_selector)
+
+#=======================================================================================
+# define algorithm:
+# Dummy algorithm
+
+from bob.pad.base.algorithm import Algorithm
+
+class DummyAlgorithm(Algorithm):
+    """An algorithm that takes the precomputed predictions and uses them for
+    scoring."""
+
+    def __init__(self, **kwargs):
+
+      super(DummyAlgorithm, self).__init__(
+          **kwargs)
+
+    def project(self, feature):
+      # print("feature",feature.as_array())
+      return feature.as_array().reshape(-1,1)
+
+
+    def score_for_multiple_projections(self, predictions):
+      # one node at the output
+
+      return list(predictions)
+
+    def score(self, predictions):
+      return list(predictions)
+algorithm = DummyAlgorithm(performs_projection=True,  requires_projector_training=False)
+
+