def TrainMaskRCNN():
ROOT_DIR = "C:\\Users\\jxmr\\Desktop\\ProjectIII\\OCRDataset\\Segmentation\\Data"
# Directory to save logs and trained model
MODEL_DIR = os.path.join(ROOT_DIR, "logs")
## Local path to trained weights file
#COCO_MODEL_PATH = os.path.join(ROOT_DIR, "mask_rcnn_coco.h5")
## Download COCO trained weights from Releases if needed
#if not os.path.exists(COCO_MODEL_PATH):
# utils.download_trained_weights(COCO_MODEL_PATH)
config = SegmentationConfig()
# Training dataset
dataset_train = SegmentationDataset()
dataset_train.load(501, 24889)
dataset_train.prepare()
# Validation dataset
dataset_val = SegmentationDataset()
dataset_val.load(1, 500)
dataset_val.prepare()
model = MaskRCNN(mode="training", config=config, model_dir=MODEL_DIR)
#model.load_weights(COCO_MODEL_PATH, by_name=True, exclude=["mrcnn_class_logits", "mrcnn_bbox_fc", "mrcnn_bbox", "mrcnn_mask"])
model.train(dataset_train,
dataset_val,
learning_rate=config.LEARNING_RATE,
epochs=8,
layers='all')
model_path = os.path.join(MODEL_DIR, "mask_rcnn_segmentation.h5")
model.keras_model.save_weights(model_path)
def main():
config = TronConfig()
config.display()
model = MaskRCNN(mode="training", config=config)
# Load weights 不能加载101
# model_path = "./pretrained/resnet101_weights_tf.h5"
# model.load_weights(model_path, by_name=True, exclude=[ "mrcnn_class_logits", "mrcnn_bbox_fc", "mrcnn_bbox", "mrcnn_mask"])
# model_path = "./pretrained/mask_rcnn_coco.h5"
model_path = "./pretrained/resnet50_weights_tf_dim_ordering_tf_kernels_notop.h5"
print("Loading weights ", model_path)
model.load_weights(model_path, by_name=True)
# Image Augmentation
# Right/Left flip 50% of the time
augmentation = imgaug.augmenters.Fliplr(0.5)
# Training - Stage 1
print("Training network heads")
model.train(learning_rate=config.LEARNING_RATE,
epochs=2,
layers='head',
augmentation=augmentation)
# Training - Stage 2
# Finetune layers from ResNet stage 4 and up
print("Fine tune Resnet stage 4 and up")
model.train(learning_rate=config.LEARNING_RATE,
epochs=2,
layers='4+', # OOM
augmentation=augmentation)
# Training - Stage 3
# Fine tune all layers
print("Fine tune all layers")
model.train(learning_rate=config.LEARNING_RATE / 10,
epochs=2,
layers='all',
augmentation=augmentation)
'learning_rate': config.LEARNING_RATE,
'layers_trained': 'all'}
subdir = os.path.dirname(model.log_dir)
if not os.path.isdir(subdir):
os.mkdir(subdir)
if not os.path.isdir(model.log_dir):
os.mkdir(model.log_dir)
train_meta_file = model.log_dir + '_meta.json'
with open(train_meta_file, 'w+') as f:
f.write(json.dumps(training_meta))
# Training all layers
model.train(dataset_train, dataset_val,learning_rate=config.LEARNING_RATE, epochs=100,
layers='all', augmentation=augmentation_pipeline)
# Training last layers
# Finetune layers from ResNet stage 4 and up
# model.train(dataset_train, dataset_val,
# learning_rate=config.LEARNING_RATE,
# epochs=120,
# layers='4+')
# Training only heads
# model.train(dataset_train, dataset_val,
# learning_rate=config.LEARNING_RATE / 10,
# epochs=160,
# layers='heads')
elif args.command == "evaluate":
elif init_with == "last":
# Load the last model you trained and continue training
weights_path = model.find_last()[1]
model.load_weights(weights_path, by_name=True)
elif init_with == 'pretrained':
weights_path = '../data/pretrained_model.h5'
model.load_weights(weights_path, by_name=True)
print('Loading weights from ', weights_path)
# Train the model for 75 epochs
model.train(dataset_train,
dataset_val,
learning_rate=1e-4,
epochs=25,
verbose=2,
layers='all')
model.train(dataset_train,
dataset_val,
learning_rate=1e-5,
epochs=50,
verbose=2,
layers='all')
model.train(dataset_train,
dataset_val,
learning_rate=1e-6,
epochs=75,
verbose=2,