# TODO: Set the paths to the datasets here.
# The directories that contain the images.
train_image_set_filename = [
"/home/uidn4455/Desktop/Devdatta_shared/dataset/COCO-text-2015/Synt-localization_dataset/Train/image",
]
gt_path_json = "/home/uidn4455/Desktop/Devdatta_shared/dataset/COCO-text-2015/Synt-localization_dataset/Train/gt_json"
val_image_set_filename = "/home/uidn4455/Desktop/Devdatta_shared/dataset/COCO-text-2015/Synt-localization_dataset/Vala/image"
val_gt_path_json = "/home/uidn4455/Desktop/Devdatta_shared/dataset/COCO-text-2015/Synt-localization_dataset/Vala/gt_json"
train_dataset.parse_json(
images_dirs=train_image_set_filename,
classes_file=
'/home/uidn4455/Desktop/Devdatta_shared/OCR/text_box_++_my_imp/class_name.json',
annotations_filenames=gt_path_json,
verbose=True,
ground_truth_available=True)
val_dataset.parse_json(
images_dirs=val_image_set_filename,
classes_file=
'/home/uidn4455/Desktop/Devdatta_shared/OCR/text_box_++_my_imp/class_name.json',
annotations_filenames=val_gt_path_json,
# verbose=True,
ground_truth_available=True)
# 3: Set the batch size.
batch_size = 16 # Change the batch size if you like, or if you run into GPU memory issues.
from models.keras_ssd7 import build_model
from keras_loss_function.keras_ssd_loss import SSDLoss
from keras_layers.keras_layer_AnchorBoxes import AnchorBoxes
from keras_layers.keras_layer_DecodeDetections import DecodeDetections
from keras_layers.keras_layer_DecodeDetectionsFast import DecodeDetectionsFast
import matplotlib.patches as patches
train_dataset = DataGenerator(load_images_into_memory=True,
hdf5_dataset_path=None)
train_dataset.parse_json(
images_dirs=images_dir,
annotations_filenames=[
"/home/jsearcy/Desktop/new_malaria/train/output.json"
],
ground_truth_available=True,
include_classes='all')
develop_dataset = DataGenerator(load_images_into_memory=True,
hdf5_dataset_path=None)
develop_dataset.parse_json(
images_dirs=images_dir,
annotations_filenames=[
"/home/jsearcy/Desktop/new_malaria/develop/output.json"
],
ground_truth_available=True,
include_classes='all')
def set_generator(self,
train_images_dir,
train_annotation_path,
batch_size,
val_images_dir=None,
val_annotation_path=None):
train_dataset = DataGenerator(load_images_into_memory=True,
hdf5_dataset_path=None)
train_dataset.parse_json(images_dirs=[train_images_dir],
annotations_filenames=[train_annotation_path],
ground_truth_available=True,
include_classes='all',
ret=False,
verbose=True)
train_dataset_size = train_dataset.get_dataset_size()
if self.model_name == 'ssd_7':
# Define the image processing chain.
ssd_data_augmentation = DataAugmentationConstantInputSize(
random_brightness=(-48, 48, 0.5),
random_contrast=(0.5, 1.8, 0.5),
random_saturation=(0.5, 1.8, 0.5),
random_hue=(18, 0.5),
random_flip=0.5,
random_translate=((0.03, 0.5), (0.03, 0.5), 0.5),
random_scale=(0.5, 2.0, 0.5),
n_trials_max=3,
clip_boxes=True,
overlap_criterion='area',
bounds_box_filter=(0.3, 1.0),
bounds_validator=(0.5, 1.0),
n_boxes_min=1,
background=(0, 0, 0))
# Instantiate an encoder that can encode ground truth labels into the format needed by the SSD loss function.
# The encoder constructor needs the spatial dimensions of the model's predictor layers to create the anchor boxes.
predictor_sizes = [
self.model.get_layer('classes4').output_shape[1:3],
self.model.get_layer('classes5').output_shape[1:3],
self.model.get_layer('classes6').output_shape[1:3],
self.model.get_layer('classes7').output_shape[1:3]
]
elif self.model_name == 'ssd_300':
# For the training generator:
ssd_data_augmentation = SSDDataAugmentation(
img_height=self.image_size[0],
img_width=self.image_size[1],
background=self.mean_color)
# 5: Instantiate an encoder that can encode ground truth labels into the format needed by the SSD loss function.
# The encoder constructor needs the spatial dimensions of the model's predictor layers to create the anchor boxes.
predictor_sizes = [
self.model.get_layer(
'conv4_3_norm_mbox_conf').output_shape[1:3],
self.model.get_layer('fc7_mbox_conf').output_shape[1:3],
self.model.get_layer('conv6_2_mbox_conf').output_shape[1:3],
self.model.get_layer('conv7_2_mbox_conf').output_shape[1:3],
self.model.get_layer('conv8_2_mbox_conf').output_shape[1:3],
self.model.get_layer('conv9_2_mbox_conf').output_shape[1:3]
]
elif self.model_name == 'ssd_512':
# For the training generator:
ssd_data_augmentation = SSDDataAugmentation(
img_height=self.image_size[0],
img_width=self.image_size[1],
background=self.mean_color)
# 5: Instantiate an encoder that can encode ground truth labels into the format needed by the SSD loss function.
# The encoder constructor needs the spatial dimensions of the model's predictor layers to create the anchor boxes.
predictor_sizes = [
self.model.get_layer(
'conv4_3_norm_mbox_conf').output_shape[1:3],
self.model.get_layer('fc7_mbox_conf').output_shape[1:3],
self.model.get_layer('conv6_2_mbox_conf').output_shape[1:3],
self.model.get_layer('conv7_2_mbox_conf').output_shape[1:3],
self.model.get_layer('conv8_2_mbox_conf').output_shape[1:3],
self.model.get_layer('conv9_2_mbox_conf').output_shape[1:3],
self.model.get_layer('conv10_2_mbox_conf').output_shape[1:3]
]
ssd_input_encoder = SSDInputEncoder(
img_height=self.image_size[0],
img_width=self.image_size[1],
n_classes=self.n_classes,
predictor_sizes=predictor_sizes,
scales=self.scales,
aspect_ratios_per_layer=self.aspect_ratios_per_layer,
two_boxes_for_ar1=self.two_boxes_for_ar1,
steps=self.steps,
offsets=self.offsets,
clip_boxes=self.clip_boxes,
variances=self.variances,
matching_type='multi',
pos_iou_threshold=0.5,
neg_iou_limit=0.5,
normalize_coords=self.normalize_coords)
self.generator = train_dataset.generate(
batch_size=batch_size,
shuffle=True,
transformations=[ssd_data_augmentation],
label_encoder=ssd_input_encoder,
returns={'processed_images', 'encoded_labels'},
keep_images_without_gt=False)
self.steps_per_epoch = ceil(train_dataset_size / batch_size)
if val_images_dir is not None and val_annotation_path is not None:
val_dataset = DataGenerator(load_images_into_memory=True,
hdf5_dataset_path=None)
val_dataset.parse_json(images_dirs=[val_images_dir],
annotations_filenames=[val_annotation_path],
ground_truth_available=True,
include_classes='all',
ret=False,
verbose=True)
val_dataset_size = val_dataset.get_dataset_size()
if self.model_name == 'ssd_300' or self.model_name == 'ssd_512':
# For the validation generator:
convert_to_3_channels = ConvertTo3Channels()
resize = Resize(height=self.image_size[0],
width=self.image_size[1])
transformations = [convert_to_3_channels, resize]
else:
transformations = []
self.validation_data = val_dataset.generate(
batch_size=batch_size,
shuffle=False,
transformations=transformations,
label_encoder=ssd_input_encoder,
returns={'processed_images', 'encoded_labels'},
keep_images_without_gt=False)
self.validation_steps = ceil(val_dataset_size / batch_size)
else:
self.validation_data = None
self.validation_steps = None
write_new_json(
train,
'/home/jsearcy/UO Data Science Dropbox/Public Datasets/BBBC041o/train.json',
folder='train')
write_new_json(
test,
'/home/jsearcy/UO Data Science Dropbox/Public Datasets/BBBC041o/test.json',
folder='test')
oa.aosefoeji
train_dataset = DataGenerator(load_images_into_memory=True,
hdf5_dataset_path=None)
train_dataset.parse_json(images_dirs=images_dir,
annotations_filenames=["output_train.json"],
ground_truth_available=True,
include_classes=use_list)
val_dataset = DataGenerator(load_images_into_memory=False,
hdf5_dataset_path=None)
val_dataset.parse_json(images_dirs=images_dir,
annotations_filenames=["output_test.json"],
ground_truth_available=True,
include_classes=use_list)
val_dataset.create_hdf5_dataset(file_path='val_dataset.h5',
resize=(1200, 1600),
variable_image_size=True,
verbose=True)
