def train():
(_, inputs) = load_images(DIR_INPUTS, INPUT_IMAGE_SIZE)
(_, teachers) = load_images(DIR_TEACHERS, INPUT_IMAGE_SIZE)
network = UNet(INPUT_IMAGE_SIZE)
model = network.model()
model.compile(optimizer='adam', loss=dice_coef_loss)
history = model.fit(
inputs, teachers, batch_size=BATCH_SIZE, epochs=EPOCHS, verbose=1)
model.save_weights(os.path.join(DIR_MODEL, File_MODEL))
plotLearningCurve(history)
def predict(input_dir, gpu_num=None):
(file_names, inputs) = load_images(input_dir,
INPUT_IMAGE_SHAPE,
with_normalize=WITH_NORM)
network = M2Det(INPUT_IMAGE_SHAPE, BATCH_SIZE, class_num=CLASS_NUM)
priors = network.get_prior_boxes()
bbox_util = BBoxUtility(CLASS_NUM, priors)
if isinstance(gpu_num, int):
model = network.get_parallel_model(gpu_num)
else:
model = network.get_model()
# model.summary()
print('loading weghts ...')
model.load_weights(os.path.join(DIR_MODEL, FILE_MODEL))
print('... loaded')
#"""
print('predicting ...')
preds = model.predict(inputs, BATCH_SIZE)
print('... predicted')
print('result saveing ...')
pred_pbox = preds[0, :, -8:]
results = bbox_util.detection_out(preds)
image_data = __outputs_to_image_data(inputs, results, file_names)
save_images(DIR_OUTPUTS,
image_data,
file_names,
with_unnormalize=WITH_NORM)
print('... finish .')
def predict(input_dir, gpu_num=None):
h, w, c = INPUT_IMAGE_SHAPE
org_h, org_w = h - (PADDING * 2), w - (PADDING * 2)
(file_names, inputs) = load_images(input_dir, (org_h, org_w, c))
inputs = np.pad(inputs, [(0, 0), (PADDING, PADDING), (PADDING, PADDING),
(0, 0)],
'constant',
constant_values=0)
network = UNet(INPUT_IMAGE_SHAPE, CLASS_NUM)
model = network.model()
plot_model(model, to_file='../model_plot.png')
model.summary()
if isinstance(gpu_num, int):
model = multi_gpu_model(model, gpus=gpu_num)
model.load_weights(os.path.join(DIR_MODEL, File_MODEL))
print('predicting ...')
preds = model.predict(inputs, BATCH_SIZE)
print('... predicted')
print('output saveing ...')
preds = preds[:, PADDING:org_h + PADDING, PADDING:org_w + PADDING, :]
save_images(DIR_OUTPUTS, preds, file_names)
print('... saved')
def predict(input_dir):
(file_names, inputs) = load_images(input_dir, INPUT_IMAGE_SIZE)
network = UNet(INPUT_IMAGE_SIZE)
model = network.model()
model.load_weights(os.path.join(DIR_MODEL, File_MODEL))
preds = model.predict(inputs, BATCH_SIZE)
save_images(DIR_OUTPUTS, preds, file_names)
def predict(input_dir, gpu_num=None):
(file_names, inputs) = load_images(input_dir, INPUT_IMAGE_SHAPE)
network = UNetPP(INPUT_IMAGE_SHAPE, start_filter=START_FILTER, depth=DEPTH, class_num=CLASS_NUM)
if isinstance(gpu_num, int):
model = network.get_parallel_model(gpu_num)
else:
model = network.get_model()
print('loading weghts ...')
model.load_weights(os.path.join(DIR_MODEL, File_MODEL))
print('... loaded')
print('predicting ...')
preds = model.predict(inputs, BATCH_SIZE)
print('... predicted')
print('result saveing ...')
save_images(DIR_OUTPUTS, preds, file_names)
print('... finish .')
def predict(input_dir, gpu_num=None):
h, w, c = INPUT_IMAGE_SHAPE
org_h, org_w = h - (PADDING * 2), w - (PADDING * 2)
(file_names, inputs) = load_images(input_dir, (org_h, org_w, c))
inputs = np.pad(inputs, [(0, 0), (PADDING, PADDING), (PADDING, PADDING), (0, 0)], 'constant', constant_values=0)
network = DeepUNet(INPUT_IMAGE_SHAPE, internal_filter=INTERNAL_FILTER, depth=DEPTH, class_num=CLASS_NUM)
if isinstance(gpu_num, int):
model = network.get_parallel_model(gpu_num)
else:
model = network.get_model()
# model.summary()
print('loading weghts ...')
model.load_weights(os.path.join(DIR_MODEL, File_MODEL))
print('... loaded')
print('predicting ...')
preds = model.predict(inputs, BATCH_SIZE)
print('... predicted')
print('result saveing ...')
preds = preds[:, PADDING:org_h+PADDING, PADDING:org_w+PADDING, :]
save_images(DIR_OUTPUTS, preds, file_names)
print('... finish .')
# model_name = "third"
# model_name = "fourthx"
# x_train = cifar_data[0][0]
# y_train = cifar_data[0][1]
#
# x_eval = cifar_data[1][0]
# y_eval = cifar_data[1][1]
# model = trainer.trainModel(model, classes, batch_size, epochs, model_name, x_train, y_train)
# trainer.evalModel(model, classes, x_eval, y_eval)
model = transferedModel()
model_name = "transfered"
flowers_data = load_images("flowers images", [".jpg"], 11 - 1, 11)
# model = load_model('models_checkpoint/transfered.hdf5')
train_flowers_x = np.array(flowers_data[0][:int(len(flowers_data[0]) * .80)])
train_flowers_y = np.array(flowers_data[1][:int(len(flowers_data[1]) * .80)])
eval_flowers_x = np.array(
flowers_data[0][int(len(flowers_data[0]) *
.80):int(len(flowers_data[0]) * 1)])
eval_flowers_y = np.array(
flowers_data[1][int(len(flowers_data[1]) *
.80):int(len(flowers_data[1]) * 1)])
train_x = np.vstack((train_flowers_x, cifar_data[0][0]))
train_y = np.vstack((train_flowers_y, cifar_data[0][1]))
eval_x = np.vstack((eval_flowers_x, cifar_data[1][0]))