def __init__(self,
patch_size=512,
saved_model_path='./model_logs/release_places2_256'):
'''
Saved model weights url:
https://drive.google.com/drive/folders/1y7Irxm3HSHGvp546hZdAZwuNmhLUVcjO
'''
self.FLAGS = ng.Config('inpaint.yml')
self.model = InpaintCAModel()
self.checkpoint_dir = saved_model_path
self.patch_size = patch_size
self.sess_config = tf.ConfigProto()
self.sess_config.gpu_options.allow_growth = True
self._inpaint_input_placeholder = tf.placeholder(
shape=(1, self.patch_size, self.patch_size * 2, 3),
dtype=tf.float32)
self.sess = tf.Session(config=self.sess_config)
output = self.model.build_server_graph(self.FLAGS,
self._inpaint_input_placeholder,
reuse=tf.AUTO_REUSE)
output = (output + 1.) * 127.5
output = tf.reverse(output, [-1])
output = tf.saturate_cast(output, tf.uint8)
self._cached_inpaint_output = output
self.load_model()
def test_save_and_run_model(tmpdir):
checkpoint_dir = tmpdir.join('ckpt') # a 'local' object.
with tf.Graph().as_default():
# Setup.
# Create model.
model = deep_fill_model.InpaintCAModel()
dummy_img = tf.zeros([batch_size, 256, 256, 3], tf.float32)
dummy_mask = tf.ones([batch_size, 256, 256, 1], tf.float32)
stage_1, stage_2, offset_flow = model.build_inpaint_net(
x=dummy_img, mask=dummy_mask, training=False, name=net_name)
with tf.Session() as sess:
# Initialize variables.
sess.run(tf.global_variables_initializer())
# Test.
# 1. Save random model. There should be no errors.
global_vars = tf.global_variables()
saver = tf.train.Saver(global_vars)
saver.save(sess, str(checkpoint_dir))
# 2. Use random model model to fill.
#input_image_path = './test/resources/case1_input.png'
#mask_path = './test/resources/case1_mask.png'
input_image_path = './waseda_fill/statue1.jpg'
mask_path = './waseda_fill/statue1_mask2.png'
out = deep_fill_app.fill(
image_path=input_image_path,
mask_path=mask_path,
checkpoint_dir=str(checkpoint_dir),
FLAGS=ng.Config('./test/resources/inpaint_test.yml'))
reference_out = str(tmpdir.join('case1_filled_using_untrained_model.png'))
cv2.imwrite('waseda_rand.png', out[0][:, :, ::-1])
(score, diff) = cv2.compare_ssim(out, reference_out, full=True)
assert score > 0.9
def deepfill_inpaint(basedata, imagepath, maskinfo, status, checkpointdir,
inputimgpath, outputpath):
# basedata = BaseData()
FLAGS = ng.Config(join(basedata.DEEPFILL_BASE_DIR, 'inpaint.yml'))
# FLAGS = ng.Config('./inpaint.yml')
# FLAGS = ng.Config('/home/zzy/work/dnnii_web/dnnii_web/App/deepfill/inpaint.yml')
# ng.get_gpus(1)
# args, unknown = parser.parse_known_args()
model = InpaintCAModel()
image = cv2.imread(imagepath)
h, w, _ = image.shape
if status == 0:
mask = np.zeros((h, w, 3)).astype(np.uint8)
for rect in maskinfo:
mask[rect[1]:rect[1] + rect[3], rect[0]:rect[0] + rect[2], :] = 255
else:
mask = cv2.imread(maskinfo)
mask = cv2.resize(mask, (w, h), fx=0.5, fy=0.5)
assert image.shape == mask.shape
#把原始图片划分成grid*grid个格子区域，'//'表示向下取整的除法
grid = 8
image = image[:h // grid * grid, :w // grid * grid, :]
mask = mask[:h // grid * grid, :w // grid * grid, :]
print('Shape of image: {}'.format(image.shape))
inputimage = image * ((255 - mask) // 255) + mask
cv2.imwrite(inputimgpath, inputimage.astype(np.uint8))
image = np.expand_dims(image, 0)
mask = np.expand_dims(mask, 0)
input_image = np.concatenate([image, mask], axis=2)
sess_config = tf.ConfigProto()
sess_config.gpu_options.allow_growth = True
deepfill_graph = tf.Graph()
with tf.Session(config=sess_config, graph=deepfill_graph) as deepfill_sess:
input_image = tf.constant(input_image, dtype=tf.float32)
output = model.build_server_graph(FLAGS, input_image)
output = (output + 1.) * 127.5
output = tf.reverse(output, [-1])
output = tf.saturate_cast(output, tf.uint8)
# load pretrained model
vars_list = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)
assign_ops = []
for var in vars_list:
vname = var.name
from_name = vname
var_value = tf.contrib.framework.load_variable(
checkpointdir, from_name)
assign_ops.append(tf.assign(var, var_value))
deepfill_sess.run(assign_ops)
print('deepfill Model loaded.')
result = deepfill_sess.run(output)
cv2.imwrite(outputpath, result[0][:, :, ::-1])
deepfill_sess.close()
def Output(img_in, mask_in, img_out):
print("import from test:", img_in, mask_in, img_out)
#if __name__ == "__main__":
FLAGS = ng.Config('inpaint.yml')
# ng.get_gpus(1)
#args, unknown = parser.parse_known_args()
model = InpaintCAModel()
#image = cv2.imread(args.image)
image = cv2.imread("examples/places356/" + img_in)
#mask = cv2.imread(args.mask)
mask = cv2.imread("places356_mask/" + mask_in)
# mask = cv2.resize(mask, (0,0), fx=0.5, fy=0.5)
assert image.shape == mask.shape
h, w, _ = image.shape
grid = 8
image = image[:h // grid * grid, :w // grid * grid, :]
mask = mask[:h // grid * grid, :w // grid * grid, :]
print('Shape of image: {}'.format(image.shape))
image = np.expand_dims(image, 0)
mask = np.expand_dims(mask, 0)
input_image = np.concatenate([image, mask], axis=2)
sess_config = tf.ConfigProto()
sess_config.gpu_options.allow_growth = True
with tf.Session(config=sess_config) as sess:
input_image = tf.constant(input_image, dtype=tf.float32)
output = model.build_server_graph(FLAGS, input_image)
output = (output + 1.) * 127.5
output = tf.reverse(output, [-1])
output = tf.saturate_cast(output, tf.uint8)
# load pretrained model
vars_list = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)
assign_ops = []
for var in vars_list:
vname = var.name
from_name = vname
#var_value = tf.contrib.framework.load_variable(args.checkpoint_dir, from_name)
var_value = tf.contrib.framework.load_variable(
"model_logs/release_places2_256", from_name)
assign_ops.append(tf.assign(var, var_value))
sess.run(assign_ops)
print('Model loaded.')
result = sess.run(output)
#cv2.imwrite(args.output, result[0][:, :, ::-1])
#cv2.imshow("result", result[0][:, :, ::-1])
cv2.imwrite("examples/places356/" + img_out, result[0][:, :, ::-1])
show1 = cv2.imread("examples/places356/" + img_in)
show2 = cv2.imread("examples/places356/" + img_out)
show = np.hstack([show1, show2])
cv2.imshow("result", show)
def inpaint(arg_image_dir, arg_mask_dir, arg_checkpoint_dir, arg_output_dir):
tf.reset_default_graph()
FLAGS = ng.Config('inpaint.yml')
# ng.get_gpus(1)
model = InpaintCAModel()
for arg_image in os.listdir(arg_image_dir):
arg_mask = arg_image # assume the mask has the same name as the image
if os.path.exists(arg_output_dir + arg_image):
print("note |", arg_image, "already inpainted.")
continue
if os.path.exists(arg_image_dir +
arg_image) and os.path.exists(arg_mask_dir +
arg_mask):
pass
else:
continue
image = cv2.imread(os.path.join(arg_image_dir, arg_image))
mask = cv2.imread(os.path.join(arg_mask_dir, arg_mask))
name = arg_image
assert image.shape == mask.shape
h, w, _ = image.shape
grid = 8
image = image[:h // grid * grid, :w // grid * grid, :]
mask = mask[:h // grid * grid, :w // grid * grid, :]
print('Shape of image: {}'.format(image.shape))
image = np.expand_dims(image, 0)
mask = np.expand_dims(mask, 0)
input_image = np.concatenate([image, mask], axis=2)
tf.reset_default_graph()
sess_config = tf.ConfigProto()
sess_config.gpu_options.allow_growth = True
with tf.Session(config=sess_config) as sess:
input_image = tf.constant(input_image, dtype=tf.float32)
output = model.build_server_graph(FLAGS, input_image)
output = (output + 1.) * 127.5
output = tf.reverse(output, [-1])
output = tf.saturate_cast(output, tf.uint8)
# load pretrained model
vars_list = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)
assign_ops = []
for var in vars_list:
vname = var.name
from_name = vname
var_value = tf.contrib.framework.load_variable(
arg_checkpoint_dir, from_name)
assign_ops.append(tf.assign(var, var_value))
sess.run(assign_ops)
print('Model loaded.')
result = sess.run(output)
cv2.imwrite(arg_output_dir + name, result[0][:, :, ::-1])
def __init__(self,
checkpoint_dir=None,
get_checkpoint_dir=download_radish,
config=abspath('inpaint.yml'),
max_size=(1024, 1024)):
self.FLAGS = ng.Config(config)
self.max_size = max_size
if checkpoint_dir is None:
checkpoint_dir = get_checkpoint_dir()
self.checkpoint_dir = checkpoint_dir
self.model_loaded = False
def inpaintP(name):
CUR_DIR = os.path.join(WORK_DIR, name)
TMASK_DIR = os.path.join(WORK_DIR, name + "//tmask")
INPAINT_MODEL_PATH = os.path.join(ROOT_DIR,
"model_logs/release_places2_256")
FLAGS = ng.Config('inpaint.yml')
model = InpaintCAModel()
image = cv.imread(os.path.join(CUR_DIR, f"{name}.png"))
mask = cv.imread(os.path.join(TMASK_DIR, "mask.png"))
filename = f'4#_{name}.png'
assert image.shape == mask.shape
h, w, _ = image.shape
grid = 8
image = image[:h // grid * grid, :w // grid * grid, :]
mask = mask[:h // grid * grid, :w // grid * grid, :]
print('Shape of image: {}'.format(image.shape))
image = np.expand_dims(image, 0)
mask = np.expand_dims(mask, 0)
input_image = np.concatenate([image, mask], axis=2)
sess_config = tf.ConfigProto()
sess_config.gpu_options.per_process_gpu_memory_fraction = 0.5
tf2 = tf.Graph()
with tf2.as_default():
with tf.Session(config=sess_config) as sess:
input_image = tf.constant(input_image, dtype=tf.float32)
output = model.build_server_graph(FLAGS, input_image)
output = (output + 1.) * 127.5
output = tf.reverse(output, [-1])
output = tf.saturate_cast(output, tf.uint8)
# load pretrained model
vars_list = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)
assign_ops = []
for var in vars_list:
vname = var.name
from_name = vname
var_value = tf.contrib.framework.load_variable(
INPAINT_MODEL_PATH, from_name)
assign_ops.append(tf.assign(var, var_value))
sess.run(assign_ops)
print('Model loaded.')
result = sess.run(output)
cv.imwrite(os.path.join(CUR_DIR, filename), result[0][:, :, ::-1])
print('Image has been made')
return 0
def generate_counterfactual(image_fp,
mask_fp,
output_fp,
checkpoint_dir,
model_id=None):
try:
FLAGS = ng.Config('config/inpaint.yml')
except AssertionError:
raise ValueError('check directory above')
# ng.get_gpus(1)
# args, unknown = parser.parse_known_args()
model = InpaintCAModel()
image = cv2.imread(image_fp)
mask = cv2.imread(mask_fp)
# mask = cv2.resize(mask, (0,0), fx=0.5, fy=0.5)
assert image.shape == mask.shape
h, w, _ = image.shape
grid = 8
image = image[:h // grid * grid, :w // grid * grid, :]
mask = mask[:h // grid * grid, :w // grid * grid, :]
image = np.expand_dims(image, 0)
mask = np.expand_dims(mask, 0)
input_image = np.concatenate([image, mask], axis=2)
sess_config = tf.ConfigProto()
sess_config.gpu_options.allow_growth = True
with tf.Session(config=sess_config) as sess:
input_image = tf.constant(input_image, dtype=tf.float32)
output = model.build_server_graph(FLAGS, input_image)
output = (output + 1.) * 127.5
output = tf.reverse(output, [-1])
output = tf.saturate_cast(output, tf.uint8)
# load pretrained model
vars_list = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)
assign_ops = []
for var in vars_list:
vname = var.name
from_name = vname
var_value = tf.contrib.framework.load_variable(
checkpoint_dir, from_name)
assign_ops.append(tf.assign(var, var_value))
sess.run(assign_ops)
print('Model loaded.')
result = sess.run(output)
cv2.imwrite(output_fp, result[0][:, :, ::-1])
print(f'IMAGE WROTE TO {output_fp}\n\n\n')
tf.reset_default_graph()
def pixel_fill(image, mask):
import tensorflow as tf2
import neuralgym as ng
from inpaint_model import InpaintCAModel
if image.ndim > mask.ndim:
mask = np.dstack([mask] * image.shape[2])
assert image.shape == mask.shape
model = InpaintCAModel()
FLAGS = ng.Config('inpaint.yml')
h, w, _ = image.shape
grid = 8
image = image[:h // grid * grid, :w // grid * grid, :]
mask = mask[:h // grid * grid, :w // grid * grid, :]
print('Shape of image: {}'.format(image.shape))
image = np.expand_dims(image, 0)
mask = np.expand_dims(mask, 0)
input_image = np.concatenate([image, mask], axis=2)
sess_config = tf2.ConfigProto()
sess_config.gpu_options.allow_growth = True
with tf2.Session(config=sess_config) as sess:
input_image = tf2.constant(input_image, dtype=tf2.float32)
output = model.build_server_graph(FLAGS, input_image)
output = (output + 1.) * 127.5
output = tf2.reverse(output, [-1])
output = tf2.saturate_cast(output, tf2.uint8)
# load pretrained model
vars_list = tf2.get_collection(tf2.GraphKeys.GLOBAL_VARIABLES)
assign_ops = []
print("checkpoint_dir = ", checkpoint_dir)
for var in vars_list:
vname = var.name
from_name = vname
if "inpaint_net" in var.name: # or else is going to mix with mrcnn
var_value = tf2.contrib.framework.load_variable(
checkpoint_dir, from_name)
assign_ops.append(tf2.assign(var, var_value))
sess.run(assign_ops)
print('Model loaded.' * 10)
result = sess.run(output)
sess.close()
tf2.reset_default_graph()
return result[0][:, :, ::-1]
def build_server_graph(self, batch_data, reuse=False, is_training=False):
"""
"""
# generate mask, 1 represents masked point
batch_raw, masks_raw = tf.split(batch_data, 2, axis=2)
masks = tf.cast(masks_raw[0:1, :, :, 0:1] > 127.5, tf.float32)
batch_pos = batch_raw / 127.5 - 1.
batch_incomplete = batch_pos * (1. - masks)
# inpaint
config = ng.Config('inpaint.yml')
x1, x2, flow = self.build_inpaint_net(
batch_incomplete, masks, reuse=reuse, training=is_training,
config=config)
batch_predict = x2
# apply mask and reconstruct
batch_complete = batch_predict*masks + batch_incomplete*(1-masks)
return batch_complete
def inpainting_api(image, mask):
FLAGS = ng.Config('inpaint.yml')
tf.reset_default_graph()
model = InpaintCAModel()
# image = cv2.imread(img_path)
# mask = cv2.imread(mask_path)
# cv2.imwrite('new.png', image - mask)
# mask = cv2.resize(mask, (0,0), fx=0.5, fy=0.5)
assert image.shape == mask.shape
image = crop(image)
mask = crop(mask)
print('Shape of image: {}'.format(image.shape))
image = np.expand_dims(image, 0)
mask = np.expand_dims(mask, 0)
input_image = np.concatenate([image, mask], axis=2)
sess_config = tf.ConfigProto()
sess_config.gpu_options.allow_growth = True
with tf.Session(config=sess_config) as sess:
input_image = tf.constant(input_image, dtype=tf.float32)
output = model.build_server_graph(FLAGS, input_image)
output = (output + 1.) * 127.5
output = tf.reverse(output, [-1])
output = tf.saturate_cast(output, tf.uint8)
# load pretrained model
vars_list = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)
assign_ops = []
for var in vars_list:
vname = var.name
from_name = vname
var_value = tf.contrib.framework.load_variable(
'./model_logs/inpaint', from_name)
assign_ops.append(tf.assign(var, var_value))
sess.run(assign_ops)
print('Model loaded.')
result = sess.run(output)
return result[0][:, :, ::-1]
def __init__(self,
checkpoint_dir,
use_gpu=False,
preview_socket_name=None):
self.video_creator = VideoCreator(fps=5)
if preview_socket_name is not None:
self.preview_socket = socket.socket(socket.AF_UNIX,
socket.SOCK_STREAM)
self.preview_socket.connect(preview_socket_name)
self.recording = False
FLAGS = ng.Config('generative_inpainting/inpaint.yml')
if use_gpu:
ng.get_gpus(1)
sess_config = tf.compat.v1.ConfigProto()
sess_config.gpu_options.allow_growth = True
self.sess = tf.compat.v1.Session(config=sess_config)
model = InpaintCAModel()
self.input_image_ph = tf.compat.v1.placeholder(tf.float32,
shape=(1, 640, 480 * 2,
3))
output = model.build_server_graph(FLAGS, self.input_image_ph)
output = (output + 1.) * 127.5
output = tf.reverse(output, [-1])
self.output = tf.saturate_cast(output, tf.uint8)
vars_list = tf.compat.v1.get_collection(
tf.compat.v1.GraphKeys.GLOBAL_VARIABLES)
assign_ops = []
for var in vars_list:
vname = var.name
from_name = vname
var_value = tf.contrib.framework.load_variable(
checkpoint_dir, from_name)
assign_ops.append(tf.compat.v1.assign(var, var_value))
self.sess.run(assign_ops)
self.on_image_received(
ThermalImage(0, Image.new('RGB', (480, 640), color='black'),
np.full((480 * 640), False)))
def inpaintP(name):
import os
import time
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
import cv2 as cv
import numpy as np
import tensorflow as tf
tf.compat.v1.logging.set_verbosity("ERROR")
import neuralgym as ng
from inpaint_model import InpaintCAModel
ROOT_DIR = os.path.abspath("./")
WAIT_DIR = os.path.abspath("./waiting")
WORK_DIR = os.path.abspath("./workspace")
CUR_DIR = os.path.join(WORK_DIR, name)
TMASK_DIR = os.path.join(WORK_DIR, name+"//tmask")
INPAINT_MODEL_PATH = os.path.join(ROOT_DIR, "model_logs/release_places2_256")
FLAGS = ng.Config('inpaint.yml')
model = InpaintCAModel()
image = cv.imread(os.path.join(CUR_DIR, f"{name}.png"))
mask = cv.imread(os.path.join(TMASK_DIR, "mask.png"))
filename = f'4#_{name}.png'
assert image.shape == mask.shape
h, w, _ = image.shape
grid = 8
image = image[:h//grid*grid, :w//grid*grid, :]
mask = mask[:h//grid*grid, :w//grid*grid, :]
print('Shape of image: {}'.format(image.shape))
image = np.expand_dims(image, 0)
mask = np.expand_dims(mask, 0)
input_image = np.concatenate([image, mask], axis=2)
sess_config = tf.ConfigProto()
sess_config.gpu_options.per_process_gpu_memory_fraction = 0.5
with tf.Session(config=sess_config) as sess:
input_image = tf.constant(input_image, dtype=tf.float32)
output = model.build_server_graph(FLAGS, input_image)
output = (output + 1.) * 127.5
output = tf.reverse(output, [-1])
output = tf.saturate_cast(output, tf.uint8)
# load pretrained model
vars_list = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)
assign_ops = []
for var in vars_list:
vname = var.name
from_name = vname
var_value = tf.contrib.framework.load_variable(INPAINT_MODEL_PATH, from_name)
assign_ops.append(tf.assign(var, var_value))
sess.run(assign_ops)
print('Model loaded.')
result = sess.run(output)
cv.imwrite(os.path.join(CUR_DIR, filename), result[0][:, :, ::-1])
summary=True,
reuse=True)
else:
_, _, losses = model.build_graph_with_losses(images,
config,
reuse=True)
if loss_type == 'g':
return losses['g_loss']
elif loss_type == 'd':
return losses['d_loss']
else:
raise ValueError('loss type is not supported.')
if __name__ == "__main__":
config = ng.Config('inpaint.yml')
if config.GPU_ID != -1:
ng.set_gpus(config.GPU_ID)
else:
ng.get_gpus(config.NUM_GPUS)
# training data
with open(config.DATA_FLIST[config.DATASET][0]) as f:
fnames = f.read().splitlines()
data = ng.data.DataFromFNames(fnames,
config.IMG_SHAPES,
random_crop=config.RANDOM_CROP)
images = data.data_pipeline(config.BATCH_SIZE)
# main model
model = InpaintCAModel()
def __init__(self, checkpoint_dir: os.PathLike,
ng_config_path: os.PathLike) -> None:
self.checkpoint_dir = checkpoint_dir
self.FLAGS = ng.Config(ng_config_path)
'--image_height', default=256, type=int,
help='The height of images should be defined, otherwise batch mode is not'
' supported.')
parser.add_argument(
'--image_width', default=256, type=int,
help='The width of images should be defined, otherwise batch mode is not'
' supported.')
parser.add_argument(
'--checkpoint_dir', default='', type=str,
help='The directory of tensorflow checkpoint.')
parser.add_argument(
'--outlist', default='', type=str,
help='The directory of putting out image.')
if __name__ == "__main__":
FLAGS = ng.Config('./inpaint_dem.yml')
ng.get_gpus(1)
# os.environ['CUDA_VISIBLE_DEVICES'] =''
args = parser.parse_args()
sess_config = tf.ConfigProto()
sess_config.gpu_options.allow_growth = True
sess = tf.Session(config=sess_config)
model = InpaintCAModel()
input_image_ph = tf.placeholder(
tf.float32, shape=(1, args.image_height, args.image_width*2, 1))
output = model.build_server_graph(FLAGS, input_image_ph)
output = tf.reverse(output, [-1])
vars_list = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)
assign_ops = []
_, _, losses = model.build_graph_with_losses(images,
config,
reuse=True)
if loss_type == 'g':
return losses['g_loss']
elif loss_type == 'd':
return losses['d_loss']
else:
raise ValueError('loss type is not supported.')
if __name__ == "__main__":
parser = ArgumentParser()
parser.add_argument('config', default='inpaint.yml')
args = parser.parse_args()
config = ng.Config(args.config)
if config.GPU_ID != -1:
ng.set_gpus(config.GPU_ID)
else:
ng.get_gpus(config.NUM_GPUS, dedicated=False)
# training data
with open(config.DATA_FLIST[config.DATASET][0]) as f:
fnames = f.read().splitlines()
data = ng.data.DataFromFNames(fnames,
config.IMG_SHAPES,
random_crop=config.RANDOM_CROP,
random_flip=config.RANDOM_FLIP,
gamma=config.GAMMA,
exposure=config.EXPOSURE)
images = data.data_pipeline(config.BATCH_SIZE)
# main model
else:
_, _, losses = model.build_graph_with_losses(images,
masks,
guides,
config,
reuse=True)
if loss_type == 'g':
return losses['g_loss']
elif loss_type == 'd':
return losses['d_loss']
else:
raise ValueError('loss type is not supported.')
if __name__ == "__main__":
config = ng.Config(sys.argv[1])
if config.GPU_ID != -1:
ng.set_gpus(config.GPU_ID)
else:
ng.get_gpus(config.NUM_GPUS)
# training data
# Image Data
with open(config.DATA_FLIST[config.DATASET][0]) as f:
fnames = f.read().splitlines()
# # Mask Data
if config.MASKFROMFILE:
with open(config.DATA_FLIST[config.MASKDATASET][0]) as f:
mask_fnames = f.read().splitlines()
data_mask_data = DataMaskFromFNames(list(zip(
fnames, mask_fnames)), [config.IMG_SHAPES, config.MASK_SHAPES],
'loss_type': 'g'
},
spe=config.TRAIN_SPE,
max_iters=config.MAX_ITERS,
log_dir=log_prefix,
)
trainer.add_callbacks([
ng.callbacks.WeightsViewer(),
# ng.callbacks.ModelRestorer(trainer.context['saver'], dump_prefix=config.RESTORE_PREFIX, optimistic=True),
discriminator_training_callback,
ng.callbacks.ModelSaver(config.TRAIN_SPE, trainer.context['saver'],
log_prefix + '/snap'),
ng.callbacks.SummaryWriter((config.VAL_PSTEPS // 1),
trainer.context['summary_writer'],
tf.summary.merge_all()),
])
trainer.train()
if __name__ == "__main__":
config = ng.Config('progressive_gan.yml')
if config.GPU_ID != -1:
ng.set_gpus(config.GPU_ID)
else:
ng.get_gpus(config.NUM_GPUS)
np.random.seed(config.RANDOM_SEED)
eval(config.TRAIN.func + '(config)')
import cv2
import numpy as np
import tensorflow as tf
import neuralgym as ng
from .inpaint_model import InpaintCAModel
checkpoint_dir = 'generative_inpainting/models'
FLAGS = ng.Config('generative_inpainting/inpaint.yml')
def run_fill(file_test, file_mask):
model = InpaintCAModel()
image = cv2.imread(file_test)
mask = cv2.imread(file_mask)
h, w, _ = image.shape
grid = 8
image = image[:h // grid * grid, :w // grid * grid, :]
mask = mask[:h // grid * grid, :w // grid * grid, :]
image = np.expand_dims(image, 0)
mask = np.expand_dims(mask, 0)
input_image = np.concatenate([image, mask], axis=2)
sess_config = tf.ConfigProto()
sess_config.gpu_options.allow_growth = True
with tf.Session(config=sess_config) as sess:
input_image = tf.constant(input_image, dtype=tf.float32)
output = model.build_server_graph(FLAGS, input_image)
output = (output + 1.) * 127.5
vars_list = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)
assign_ops = []
idx_list = [54, 56, 58, 60, 62] # random
for idx, var in enumerate(vars_list):
vname = var.name
from_name = vname
var_value = tf.contrib.framework.load_variable(checkpoint_dir,
from_name)
assign_ops.append(tf.assign(var, var_value))
sess.run(assign_ops)
print('Model loaded.')
result = sess.run(output)
return result
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--image', default='', type=str,
help='The filename of image to be completed.')
parser.add_argument('--mask', default='', type=str,
help='The filename of mask, value 255 indicates mask.')
parser.add_argument('--output', default='output.png', type=str,
help='Where to write output.')
parser.add_argument('--checkpoint_dir', default='', type=str,
help='The directory of tensorflow checkpoint.')
FLAGS = ng.Config('./resources/inpaint.yml')
args, _ = parser.parse_known_args()
result = fill(args.image, args.mask, args.checkpoint_dir, FLAGS)
cv2.imwrite(args.output, result[0][:, :, ::-1])
reuse=True)
if loss_type == 'g':
return losses['g_loss']
elif loss_type == 'd':
return losses['d_loss']
else:
raise ValueError('loss type is not supported.')
if __name__ == "__main__":
if len(sys.argv) > 1:
yml_path = sys.argv[1]
else:
yml_path = 'inpaint.yml'
config = ng.Config(yml_path)
if config.GPU_ID != -1:
ng.set_gpus(config.GPU_ID)
else:
ng.get_gpus(config.NUM_GPUS)
# training data
with open(config.DATA_FLIST[config.DATASET][0]) as f:
fnames = f.read().splitlines()
data = ng.data.DataFromFNames(fnames,
config.IMG_SHAPES,
random_crop=config.RANDOM_CROP)
images = data.data_pipeline(config.BATCH_SIZE)
# main model
model = InpaintCAModel()
g_vars, d_vars, losses = model.build_graph_with_losses(images,
config=config)
def cagc_inp_batch():
baserecutpath = '/home/zzy/TrainData/MITPlace2Dataset/val_recut_512x680'
imagesnums = 132
checkpointdir = "./checkpoints/places2_512x680"
maskpath = "./examples/places2_680x512/wooden_mask.png"
recutoutput = './examples/recutoutput/'
recutmaskedoutput = './examples/recutmasked/'
FLAGS = ng.Config('inpaint.yml')
# ng.get_gpus(1)
# args, unknown = parser.parse_known_args()
model = InpaintCAModel()
sess_config = tf.ConfigProto()
sess_config.gpu_options.allow_growth = True
with tf.Session(config=sess_config) as sess:
for imageindex in range(1, imagesnums):
imagename = 'Places365_' + str(imageindex).zfill(8) + '.png'
maskedimagename = 'Places365_' + str(imageindex).zfill(
8) + '_masked.png'
outputimagename = 'Places365_' + str(imageindex).zfill(
8) + '_output.png'
imagepath = os.path.join(baserecutpath, imagename)
outputpath = os.path.join(recutoutput, outputimagename)
maskedimagepath = os.path.join(recutmaskedoutput, maskedimagename)
image = cv2.imread(imagepath)
mask = cv2.imread(maskpath)
h, w, _ = image.shape
mask = cv2.resize(mask, (w, h), fx=0.5, fy=0.5)
inputimage = image * ((255 - mask) // 255) + mask
cv2.imwrite(maskedimagepath, inputimage.astype(np.uint8))
assert image.shape == mask.shape
# 把原始图片划分成grid*grid个格子区域，'//'表示向下取整的除法
grid = 8
image = image[:h // grid * grid, :w // grid * grid, :]
mask = mask[:h // grid * grid, :w // grid * grid, :]
print('Shape of image: {}'.format(image.shape))
image = np.expand_dims(image, 0)
mask = np.expand_dims(mask, 0)
input_image = np.concatenate([image, mask], axis=2)
input_image = tf.constant(input_image, dtype=tf.float32)
output = model.build_server_graph(FLAGS, input_image)
output = (output + 1.) * 127.5
output = tf.reverse(output, [-1])
output = tf.saturate_cast(output, tf.uint8)
# load pretrained model
vars_list = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)
assign_ops = []
for var in vars_list:
vname = var.name
from_name = vname
var_value = tf.contrib.framework.load_variable(
checkpointdir, from_name)
assign_ops.append(tf.assign(var, var_value))
sess.run(assign_ops)
print('Model loaded.')
result = sess.run(output)
cv2.imwrite(outputpath, result[0][:, :, ::-1])
else:
_, _, losses = model.build_graph_with_losses(FLAGS,
images,
FLAGS,
reuse=True)
if loss_type == 'g':
return losses['g_loss']
elif loss_type == 'd':
return losses['d_loss']
else:
raise ValueError('loss type is not supported.')
if __name__ == "__main__":
# training data
FLAGS = ng.Config('inpaint.yml')
img_shapes = FLAGS.img_shapes
with open(FLAGS.data_flist[FLAGS.dataset][0]) as f:
fnames = f.read().splitlines()
if FLAGS.guided:
fnames = [(fname, fname[:-4] + '_edge.jpg') for fname in fnames]
img_shapes = [img_shapes, img_shapes]
data = ng.data.DataFromFNames(fnames,
img_shapes,
random_crop=FLAGS.random_crop,
nthreads=FLAGS.num_cpus_per_job)
images = data.data_pipeline(FLAGS.batch_size)
# main model
model = InpaintCAModel()
g_vars, d_vars, losses = model.build_graph_with_losses(FLAGS, images)
# validation images
def deepfillbatch(image_height, image_width, checkpoint_dir, img_mask_txt,
outputdir):
FLAGS = ng.Config('./deepfill/inpaint.yml')
ng.get_gpus(1)
# os.environ['CUDA_VISIBLE_DEVICES'] =''
sess_config = tf.ConfigProto()
sess_config.gpu_options.allow_growth = True
sess = tf.Session(config=sess_config)
model = InpaintCAModel()
input_image_ph = tf.placeholder(tf.float32,
shape=(1, image_height, image_width * 2,
3))
output = model.build_server_graph(FLAGS, input_image_ph)
output = (output + 1.) * 127.5
output = tf.reverse(output, [-1])
output = tf.saturate_cast(output, tf.uint8)
vars_list = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)
assign_ops = []
for var in vars_list:
vname = var.name
from_name = vname
var_value = tf.contrib.framework.load_variable(checkpoint_dir,
from_name)
assign_ops.append(tf.assign(var, var_value))
sess.run(assign_ops)
print('Model loaded.')
t = time.time()
with open(img_mask_txt, 'r') as f:
while True:
line = f.readline() # 整行读取数据
if not line:
break
imagepath, maskpath = line.replace('\n', '').replace('\r',
'').split(',')
print(imagepath)
print(maskpath)
if not os.path.exists(outputdir):
os.mkdir(outputdir)
outputpath = os.path.join(outputdir, os.path.basename(imagepath))
image = cv2.imread(imagepath)
mask = cv2.imread(maskpath)
print(image.shape)
print(mask.shape)
image = cv2.resize(image, (image_width, image_height))
mask = cv2.resize(mask, (image_width, image_height))
# cv2.imwrite(maskedimg, image*(1-mask/255.) + mask)
# # continue
# image = np.zeros((128, 256, 3))
# mask = np.zeros((128, 256, 3))
assert image.shape == mask.shape
h, w, _ = image.shape
grid = 4
image = image[:h // grid * grid, :w // grid * grid, :]
mask = mask[:h // grid * grid, :w // grid * grid, :]
print('Shape of image: {}'.format(image.shape))
image = np.expand_dims(image, 0)
mask = np.expand_dims(mask, 0)
input_image = np.concatenate([image, mask], axis=2)
# load pretrained model
result = sess.run(output, feed_dict={input_image_ph: input_image})
print('Processed: {}'.format(outputpath))
cv2.imwrite(outputpath, result[0][:, :, ::-1])
print('Time total: {}'.format(time.time() - t))