self.training_trainer = SupervisedTrainer
self.training_epochs = 10
self.training_batch_size = 32
# Should be in a model.py
class ImageClassifierSimple(nn.Module):
def __init__(self, config):
super().__init__()
self.config = config
self.layers = Sequential(
ImageConversion(standardize=False, to_channel_first=True),
Conv2D(kernel_size=(3, 3), filters=12), Activation("relu"),
MaxPooling2D(), BatchNormalization(),
Conv2D(kernel_size=(3, 3), filters=18), Activation("relu"),
MaxPooling2D(), BatchNormalization(),
Conv2D(kernel_size=(3, 3), filters=18), Activation("relu"),
MaxPooling2D(), BatchNormalization(),
Conv2D(kernel_size=(3, 3), filters=18), Activation("relu"),
MaxPooling2D(), BatchNormalization(), Flatten(), Dense(18),
Activation("relu"), Dense(10), Activation("softmax", dim=1))
def forward(self, image):
return self.layers(image)
# Run with parameters parsed from commandline.
# python -m deeptech.examples.mnist_custom_model --mode=train --input=Datasets --output=Results
if __name__ == "__main__":
cli.run(FashionMNISTConfig)
self.training_initial_lr = 0.001
def create_loss(self, model):
rpn_loss = DetectionLoss(anchors="rpn_anchors",
pred_boxes="rpn_deltas",
pred_class_ids="rpn_class_ids",
target_boxes="boxes",
target_class_ids="fg_bg_classes",
channel_last_gt=True,
lower_tresh=0.3,
upper_tresh=0.5,
delta_preds=not self.model_log_delta_preds,
log_delta_preds=self.model_log_delta_preds)
final_loss = DetectionLoss(anchors="final_anchors",
pred_boxes="final_deltas",
pred_class_ids="final_class_ids",
target_boxes="boxes",
target_class_ids="class_ids",
channel_last_gt=True,
lower_tresh=0.5,
upper_tresh=0.7,
delta_preds=not self.model_log_delta_preds,
log_delta_preds=self.model_log_delta_preds)
return MultiLoss(model, rpn=rpn_loss, final=final_loss)
# Run with parameters parsed from commandline.
# python -m deeptech.examples.mnist_custom_loss --mode=train --input=Datasets --output=Results
if __name__ == "__main__":
cli.run(COCOFasterRCNNConfig)