CNN初学者，最近自己在github上拿了个项目练手，问题是数据集不公开，只能自己做数据集，但是却看不懂数据集应该怎么制作。
代码如下
应该就是DAC_DATASET类中
class DAC_Dataset(RNGDataFlow):
def init(self, dataset_dir, train, all_classes):
self.images = []
if all_classes == 1:
for directory in listdir(dataset_dir):
for file in listdir(dataset_dir + '/' + directory):
if '.jpg' in file:
for c in classes:
if c[0] in directory:
label = c[1]
break
self.images.append([dataset_dir + '/' + directory + '/' + file, label])
else:
for file in listdir(dataset_dir):
if '.jpg' in file:
self.images.append([dataset_dir + '/' + file, 0])
shuffle(self.images)
if train == 0:
self.images = self.images[0:1000]
def get_data(self):
for image in self.images:
xml_name = image[0].replace('jpg','xml')
im = cv2.imread(image[0], cv2.IMREAD_COLOR)
im = cv2.resize(im, (square_size, square_size))
im = im.reshape((square_size, square_size, 3))
meta = None
if os.path.isfile(image[0].replace('jpg','xml')):
meta = xml.etree.ElementTree.parse(xml_name).getroot()
label = np.array(image[1])
bndbox = {}
bndbox['xmin'] = 0
bndbox['xmax'] = 0
bndbox['ymin'] = 0
bndbox['ymax'] = 0
if meta is not None:
obj = meta.find('object')
if obj is not None:
box = obj.find('bndbox')
if box is not None:
bndbox['xmin'] = int(box.find('xmin').text)
bndbox['xmax'] = int(box.find('xmax').text)
bndbox['ymin'] = int(box.find('ymin').text)
bndbox['ymax'] = int(box.find('ymax').text)
bndbox['xmin'] = int(bndbox['xmin']*(square_size/IMAGE_WIDTH))
bndbox['xmax'] = int(bndbox['xmax']*(square_size/IMAGE_WIDTH))
bndbox['ymin'] = int(bndbox['ymin']*(square_size/IMAGE_HEIGHT))
bndbox['ymax'] = int(bndbox['ymax']*(square_size/IMAGE_HEIGHT))
iou = np.zeros( (height_width, height_width) )
for h in range(0, height_width):
for w in range(0, height_width):
rect = {}
rect['xmin'] = int(w*down_sample_factor)
rect['xmax'] = int((w+1)*down_sample_factor)
rect['ymin'] = int(h*down_sample_factor)
rect['ymax'] = int((h+1)*down_sample_factor)
if DEMO_DATASET == 0:
if intersection(rect, bndbox) == 0.0:
iou[h,w] = 0.0
else:
iou[h,w] = 1.0
else:
if intersection(rect, bndbox) < 0.5:
iou[h,w] = 0.0
else:
iou[h,w] = 1.0
# if iou[h,w] > 0:
# cv2.rectangle(im, (int(rect['xmin']),int(rect['ymin'])), (int(rect['xmax']),int(rect['ymax'])), (0,0,iou[h,w]*255), 1)
iou = iou.reshape( (height_width, height_width, 1) )
valid = np.zeros((height_width, height_width, 4), dtype='float32')
relative_bndboxes = np.zeros((height_width, height_width, 4), dtype='float32')
for h in range(0, height_width):
for w in range(0, height_width):
if iou[h, w] > 0.0:
valid[h,w,0] = 1.0
valid[h,w,1] = 1.0
valid[h,w,2] = 1.0
valid[h,w,3] = 1.0
relative_bndboxes[h, w, 0] = bndbox['xmin'] - w*down_sample_factor
relative_bndboxes[h, w, 1] = bndbox['ymin'] - h*down_sample_factor
relative_bndboxes[h, w, 2] = bndbox['xmax'] - w*down_sample_factor
relative_bndboxes[h, w, 3] = bndbox['ymax'] - h*down_sample_factor
else:
relative_bndboxes[h, w] = np.zeros(4)
# cv2.rectangle(im, (bndbox['xmin'],bndbox['ymin']), (bndbox['xmax'],bndbox['ymax']), (255,0,0), 1)
# cv2.imshow('image', im)
# cv2.waitKey(1000)
yield [im, label, iou, valid, relative_bndboxes]
def size(self):
return len(self.images)
class Model(ModelDesc):
def _get_inputs(self):
return [InputDesc(tf.float32, [None, square_size, square_size, 3], 'input'),
InputDesc(tf.int32, [None], 'label'),
InputDesc(tf.float32, [None, height_width, height_width, 1], 'ious'),
InputDesc(tf.float32, [None, height_width, height_width, 4], 'valids'),
InputDesc(tf.float32, [None, height_width, height_width, 4], 'bndboxes')]
def _build_graph(self, inputs):
image, label, ious, valids, bndboxes = inputs
image = tf.round(image)
fw, fa, fg = get_dorefa(BITW, BITA, BITG)
old_get_variable = tf.get_variable
def monitor(x, name):
if MONITOR == 1:
return tf.Print(x, [x], message='\n\n' + name + ': ', summarize=1000, name=name)
else:
return x
def new_get_variable(v):
name = v.op.name
if not name.endswith('W') or 'conv1' in name or 'conv_obj' in name or 'conv_box' in name:
return v
else:
logger.info("Quantizing weight {}".format(v.op.name))
if MONITOR == 1:
return tf.Print(fw(v), [fw(v)], message='\n\n' + v.name + ', Quantized weights are:', summarize=100)
else:
return fw(v)
def activate(x):
if BITA == 32:
return tf.nn.relu(x)
else:
return fa(tf.nn.relu(x))
def bn_activate(name, x):
x = BatchNorm(name, x)
x = monitor(x, name + '_noact_out')
return activate(x)
def halffire(name, x, num_squeeze_filters, num_expand_3x3_filters, skip):
out_squeeze = Conv2D('squeeze_conv_' + name, x, out_channel=num_squeeze_filters, kernel_shape=1, stride=1, padding='SAME')
out_squeeze = bn_activate('bn_squeeze_' + name, out_squeeze)
out_expand_3x3 = Conv2D('expand_3x3_conv_' + name, out_squeeze, out_channel=num_expand_3x3_filters, kernel_shape=3, stride=1, padding='SAME')
out_expand_3x3 = bn_activate('bn_expand_3x3_' + name, out_expand_3x3)
if skip == 0:
return out_expand_3x3
else:
return tf.add(x, out_expand_3x3)
def halffire_noact(name, x, num_squeeze_filters, num_expand_3x3_filters):
out_squeeze = Conv2D('squeeze_conv_' + name, x, out_channel=num_squeeze_filters, kernel_shape=1, stride=1, padding='SAME')
out_squeeze = bn_activate('bn_squeeze_' + name, out_squeeze)
out_expand_3x3 = Conv2D('expand_3x3_conv_' + name, out_squeeze, out_channel=num_expand_3x3_filters, kernel_shape=3, stride=1, padding='SAME')
return out_expand_3x3
with remap_variables(new_get_variable), \
argscope([Conv2D, FullyConnected], use_bias=False, nl=tf.identity), \
argscope(BatchNorm, decay=0.9, epsilon=1e-4):
image = monitor(image, 'image_out')
l = Conv2D('conv1', image, out_channel=32, kernel_shape=3, stride=2, padding='SAME')
l = bn_activate('bn1', l)
l = monitor(l, 'conv1_out')
l = MaxPooling('pool1', l, shape=3, stride=2, padding='SAME')
l = monitor(l, 'pool1_out')
l = halffire('fire1', l, NUM_SQUEEZE_FILTERS, NUM_EXPAND_FILTERS, 0)
l = monitor(l, 'fire1_out')
l = MaxPooling('pool2', l, shape=3, stride=2, padding='SAME')
l = monitor(l, 'pool2_out')
l = halffire('fire2', l, NUM_SQUEEZE_FILTERS, NUM_EXPAND_FILTERS, 0)
l = monitor(l, 'fire2_out')
l = MaxPooling('pool3', l, shape=3, stride=2, padding='SAME')
l = monitor(l, 'pool3_out')
l = halffire('fire3', l, NUM_SQUEEZE_FILTERS, NUM_EXPAND_FILTERS, 0)
l = monitor(l, 'fire3_out')
l = halffire('fire4', l, NUM_SQUEEZE_FILTERS, NUM_EXPAND_FILTERS, 0)
l = monitor(l, 'fire4_out')
l = halffire('fire5', l, NUM_SQUEEZE_FILTERS, NUM_EXPAND_FILTERS, 0)
l = monitor(l, 'fire5_out')
l = halffire('fire6', l, NUM_SQUEEZE_FILTERS, NUM_EXPAND_FILTERS, 0)
l = monitor(l, 'fire6_out')
l = halffire('fire7', l, NUM_SQUEEZE_FILTERS, NUM_EXPAND_FILTERS, 0)
l = monitor(l, 'fire7_out')
# Classification
classify = Conv2D('conv_class', l, out_channel=12, kernel_shape=1, stride=1, padding='SAME')
classify = bn_activate('bn_class', classify)
classify = monitor(classify, 'conv_class_out')
logits = GlobalAvgPooling('pool_class', classify)
class_loss = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=logits, labels=label)
class_loss = tf.reduce_mean(class_loss, name='cross_entropy_loss')
wrong = prediction_incorrect(logits, label, 1, name='wrong-top1')
add_moving_summary(tf.reduce_mean(wrong, name='train-error-top1'))
# Object Detection
l = tf.concat([l, classify], axis=3)
objdetect = Conv2D('conv_obj', l, out_channel=1, kernel_shape=1, stride=1, padding='SAME')
objdetect = tf.identity(objdetect, name='objdetect_out')
objdetect_loss = tf.losses.hinge_loss(labels=ious, logits=objdetect)
bndbox = Conv2D('conv_box', l, out_channel=4, kernel_shape=1, stride=1, padding='SAME')
bndbox = tf.identity(bndbox, name='bndbox_out')
bndbox = tf.multiply(bndbox, valids, name='mult0')
bndbox_loss = tf.losses.mean_squared_error(labels=bndboxes, predictions=bndbox)
# weight decay on all W of fc layers
# reg_cost = regularize_cost('(fire7|conv_obj|conv_box).*/W', l2_regularizer(1e-5), name='regularize_cost')
# cost = class_loss*objdetect_loss*bndbox_loss
# cost = class_loss + objdetect_loss + bndbox_loss + reg_cost
cost = class_loss + 10*objdetect_loss + bndbox_loss
add_moving_summary(class_loss, objdetect_loss, bndbox_loss, cost)
self.cost = cost
tf.get_variable = old_get_variable
def _get_optimizer(self):
lr = tf.get_variable('learning_rate', initializer=1e-2, trainable=False)
opt = tf.train.AdamOptimizer(lr, epsilon=1e-5)
# lr = tf.get_variable('learning_rate', initializer=1e-1, trainable=False)
# opt = tf.train.MomentumOptimizer(lr, momentum=0.9)
return opt
def get_data(dataset_dir, train):
if DEMO_DATASET == 0:
all_classes = 1
else:
all_classes = 0
ds = DAC_Dataset(dataset_dir, train, all_classes)
ds = BatchData(ds, BATCH_SIZE, remainder=False)
ds = PrefetchDataZMQ(ds, nr_proc=8, hwm=6)
return ds
def get_config():
logger.auto_set_dir()
data_train = get_data(args.data, 1)
data_test = get_data(args.data, 0)
if DEMO_DATASET == 0:
return TrainConfig(
dataflow=data_train,
callbacks=[
ModelSaver(max_to_keep=10),
HumanHyperParamSetter('learning_rate'),
ScheduledHyperParamSetter('learning_rate', [(40, 0.001), (60, 0.0001), (90, 0.00001)])
,InferenceRunner(data_test,
[ScalarStats('cross_entropy_loss'),
ClassificationError('wrong-top1', 'val-error-top1')])
],
model=Model(),
max_epoch=150
)
else:
return TrainConfig(
dataflow=data_train,
callbacks=[
ModelSaver(max_to_keep=10),
HumanHyperParamSetter('learning_rate'),
ScheduledHyperParamSetter('learning_rate', [(100, 0.001), (200, 0.0001), (250, 0.00001)])
],
model=Model(),
max_epoch=300
)
def run_image(model, sess_init, image_dir):
print('Running image!')
output_names = ['objdetect_out', 'bndbox_out']
pred_config = PredictConfig(
model=model,
session_init=sess_init,
input_names=['input'],
output_names=output_names
)
predictor = OfflinePredictor(pred_config)
images = []
metas = []
for file in listdir(image_dir):
if '.jpg' in file:
images.append(file)
if '.xml' in file:
metas.append(file)
images.sort()
metas.sort()
THRESHOLD = 0
index = 0
for image in images:
meta = xml.etree.ElementTree.parse(image_dir + '/' + metas[index]).getroot()
true_bndbox = {}
true_bndbox['xmin'] = 0
true_bndbox['xmax'] = 0
true_bndbox['ymin'] = 0
true_bndbox['ymax'] = 0
if meta is not None:
obj = meta.find('object')
if obj is not None:
box = obj.find('bndbox')
if box is not None:
true_bndbox['xmin'] = int(box.find('xmin').text)
true_bndbox['xmax'] = int(box.find('xmax').text)
true_bndbox['ymin'] = int(box.find('ymin').text)
true_bndbox['ymax'] = int(box.find('ymax').text)
index += 1
im = cv2.imread(image_dir + '/' + image, cv2.IMREAD_COLOR)
im = cv2.resize(im, (square_size, square_size))
im = im.reshape((1, square_size, square_size, 3))
outputs = predictor([im])
im = cv2.imread(image_dir + '/' + image, cv2.IMREAD_COLOR)
objdetect = outputs[0]
bndboxes = outputs[1]
max_pred = -100
max_h = -1
max_w = -1
for h in range(0, objdetect.shape[1]):
for w in range(0, objdetect.shape[2]):
if objdetect[0, h, w] > max_pred:
max_pred = objdetect[0, h, w]
max_h = h
max_w = w
sum_labels= 0;
bndbox = {}
bndbox['xmin'] = 0
bndbox['ymin'] = 0
bndbox['xmax'] = 0
bndbox['ymax'] = 0
for h in range(0, objdetect.shape[1]):
for w in range(0, objdetect.shape[2]):
if (objdetect[0, h, w] > THRESHOLD and (h == max_h-1 or h == max_h or h == max_h+1) and (w == max_w-1 or w == max_w or w == max_w+1)) or (h == max_h and w == max_w):
sum_labels += 1
bndbox['xmin'] += int( (bndboxes[0,h,w,0] + w*down_sample_factor) )
bndbox['ymin'] += int( (bndboxes[0,h,w,1] + h*down_sample_factor) )
bndbox['xmax'] += int( (bndboxes[0,h,w,2] + w*down_sample_factor) )
bndbox['ymax'] += int( (bndboxes[0,h,w,3] + h*down_sample_factor) )
temp_xmin = int( (bndboxes[0,h,w,0] + w*down_sample_factor) *(IMAGE_WIDTH/square_size) )
temp_ymin = int( (bndboxes[0,h,w,1] + h*down_sample_factor) *(IMAGE_HEIGHT/square_size) )
temp_xmax = int( (bndboxes[0,h,w,2] + w*down_sample_factor) *(IMAGE_WIDTH/square_size) )
temp_ymax = int( (bndboxes[0,h,w,3] + h*down_sample_factor) *(IMAGE_HEIGHT/square_size) )
cv2.rectangle(im, (temp_xmin,temp_ymin), (temp_xmax,temp_ymax), (255,0,0), 1)
bndbox['xmin'] = int(bndbox['xmin']*(1/sum_labels))
bndbox['ymin'] = int(bndbox['ymin']*(1/sum_labels))
bndbox['xmax'] = int(bndbox['xmax']*(1/sum_labels))
bndbox['ymax'] = int(bndbox['ymax']*(1/sum_labels))
bndbox['xmin'] = int(bndbox['xmin']*(IMAGE_WIDTH/square_size))
bndbox['ymin'] = int(bndbox['ymin']*(IMAGE_HEIGHT/square_size))
bndbox['xmax'] = int(bndbox['xmax']*(IMAGE_WIDTH/square_size))
bndbox['ymax'] = int(bndbox['ymax']*(IMAGE_HEIGHT/square_size))
bndbox2 = {}
bndbox2['xmin'] = int( bndboxes[0,max_h,max_w,0] + max_w*down_sample_factor)
bndbox2['ymin'] = int( bndboxes[0,max_h,max_w,1] + max_h*down_sample_factor)
bndbox2['xmax'] = int( bndboxes[0,max_h,max_w,2] + max_w*down_sample_factor)
bndbox2['ymax'] = int( bndboxes[0,max_h,max_w,3] + max_h*down_sample_factor)
bndbox2['xmin'] = int(bndbox2['xmin']*(IMAGE_WIDTH/square_size))
bndbox2['ymin'] = int(bndbox2['ymin']*(IMAGE_HEIGHT/square_size))
bndbox2['xmax'] = int(bndbox2['xmax']*(IMAGE_WIDTH/square_size))
bndbox2['ymax'] = int(bndbox2['ymax']*(IMAGE_HEIGHT/square_size))
print('----------------------------------------')
print(str(max_h*14+max_w))
print('xmin: ' + str(bndbox2['xmin']))
print('xmax: ' + str(bndbox2['xmax']))
print('ymin: ' + str(bndbox2['ymin']))
print('ymax: ' + str(bndbox2['ymax']))
cv2.rectangle(im, (int(max_w*down_sample_factor*(IMAGE_WIDTH/square_size)),int(max_h*down_sample_factor*(IMAGE_HEIGHT/square_size))), (int((max_w+1)*down_sample_factor*(IMAGE_WIDTH/square_size)),int((max_h+1)*down_sample_factor*(IMAGE_HEIGHT/square_size))), (0,0,255), 1)
cv2.rectangle(im, (true_bndbox['xmin'], true_bndbox['ymin']), (true_bndbox['xmax'], true_bndbox['ymax']), (255,0,0), 2)
cv2.rectangle(im, (bndbox2['xmin'], bndbox2['ymin']), (bndbox2['xmax'],bndbox2['ymax']), (0,255,0), 2)
cv2.imshow('image', im)
cv2.imwrite('images_log/' + image, im)
cv2.waitKey(800)
def run_single_image(model, sess_init, image):
print('Running single image!')
if MONITOR == 1:
monitor_names = ['conv_class_out', 'image_out', 'conv1_out', 'pool1_out', 'fire1_out', 'pool2_out', 'pool3_out', 'fire5_out', 'fire6_out', 'fire7_out']
else:
monitor_names = []
output_names = ['objdetect_out', 'bndbox_out']
output_names.extend(monitor_names)
pred_config = PredictConfig(
model=model,
session_init=sess_init,
input_names=['input'],
output_names=output_names
)
predictor = OfflinePredictor(pred_config)
if REAL_IMAGE == 1:
im = cv2.imread(image, cv2.IMREAD_COLOR)
im = cv2.resize(im, (square_size, square_size))
cv2.imwrite('test_image.png', im)
im = im.reshape((1, square_size, square_size, 3))
else:
im = np.zeros((1, square_size, square_size, 3))
k = 0
for h in range(0, square_size):
for w in range(0,square_size):
for c in range (0,3):
# im[0][h][w][c] = 0
im[0][h][w][c] = k%256
k += 1
outputs = predictor([im])
objdetect = outputs[0]
bndboxes = outputs[1]
max_pred = -100
max_h = -1
max_w = -1
for h in range(0, objdetect.shape[1]):
for w in range(0, objdetect.shape[2]):
if objdetect[0, h, w] > max_pred:
max_pred = objdetect[0, h, w]
max_h = h
max_w = w
bndbox2 = {}
bndbox2['xmin'] = int( bndboxes[0,max_h,max_w,0] + max_w*down_sample_factor)
bndbox2['ymin'] = int( bndboxes[0,max_h,max_w,1] + max_h*down_sample_factor)
bndbox2['xmax'] = int( bndboxes[0,max_h,max_w,2] + max_w*down_sample_factor)
bndbox2['ymax'] = int( bndboxes[0,max_h,max_w,3] + max_h*down_sample_factor)
bndbox2['xmin'] = int(bndbox2['xmin']*(640/square_size))
bndbox2['ymin'] = int(bndbox2['ymin']*(360/square_size))
bndbox2['xmax'] = int(bndbox2['xmax']*(640/square_size))
bndbox2['ymax'] = int(bndbox2['ymax']*(360/square_size))
# im = cv2.imread(image, cv2.IMREAD_COLOR)
# cv2.rectangle(im, (bndbox2['xmin'], bndbox2['ymin']), (bndbox2['xmax'],bndbox2['ymax']), (0,255,0), 2)
# cv2.imshow('image', im)
# cv2.waitKey(2000)
print('max_h: ' + str(max_h))
print('max_w: ' + str(max_w))
print('objdetect: ' + str(objdetect))
print('bndboxes: ' + str(bndboxes[0,max_h,max_w]))
index = 2
for o in monitor_names:
print(o + ', shape: ' + str(outputs[index].shape) )
if 'image' not in o:
print(str(outputs[index]))
if len(outputs[index].shape) == 4:
file_name = o.split('/')[-1]
print('Writing file... ' + file_name)
if not os.path.exists('./log'):
os.makedirs('./log')
with open('./log/' + file_name + '.log', 'w') as f:
for sample in range(0, outputs[index].shape[0]):
for h in range(0, outputs[index].shape[1]):
for w in range(0, outputs[index].shape[2]):
res = ''
for c in range(0, outputs[index].shape[3]):
if 'image' in file_name:
res = hexFromInt( int(outputs[index][sample, h, w, c]), 8 ) + '_' + res
elif 'noact' in file_name:
temp = (2**FACTOR_SCALE_BITS)*outputs[index][sample, h, w, c]
res = hexFromInt( int(temp), 32 ) + '_' + res
else:
res = hexFromInt( int(outputs[index][sample, h, w, c]), BITA) + '_' + res
f.write('0x' + res + '\n')
index += 1
def dump_weights(meta, model, output):
fw, fa, fg = get_dorefa(BITW, BITA, BITG)
with tf.Graph().as_default() as G:
tf.train.import_meta_graph(meta)
init = get_model_loader(model)
sess = tf.Session(config=tf.ConfigProto(allow_soft_placement=True))
sess.run(tf.global_variables_initializer())
init.init(sess)
with sess.as_default():
if output:
if output.endswith('npy') or output.endswith('npz'):
varmanip.dump_session_params(output)
else:
var = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES)
var.extend(tf.get_collection(tf.GraphKeys.MODEL_VARIABLES))
var_dict = {}
for v in var:
name = varmanip.get_savename_from_varname(v.name)
var_dict[name] = v
logger.info("Variables to dump:")
logger.info(", ".join(var_dict.keys()))
saver = tf.train.Saver(
var_list=var_dict,
write_version=tf.train.SaverDef.V2)
saver.save(sess, output, write_meta_graph=False)
network_model = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES)
network_model.extend(tf.get_collection(tf.GraphKeys.MODEL_VARIABLES))
target_frequency = 200000000
target_FMpS = 300
non_quantized_layers = ['conv1/Conv2D', 'conv_obj/Conv2D', 'conv_box/Conv2D']
json_out, layers_list, max_cycles = generateLayers(sess, BITA, BITW, non_quantized_layers, target_frequency, target_FMpS)
achieved_FMpS = target_frequency/max_cycles
if DEMO_DATASET == 0:
generateConfig(layers_list, 'halfsqueezenet-config.h')
genereateHLSparams(layers_list, network_model, 'halfsqueezenet-params.h', fw)
else:
generateConfig(layers_list, 'halfsqueezenet-config_demo.h')
genereateHLSparams(layers_list, network_model, 'halfsqueezenet-params_demo.h', fw)
print('|---------------------------------------------------------|')
print('target_FMpS: ' + str(target_FMpS) )
print('achieved_FMpS: ' + str(achieved_FMpS) )
if name == '__main__':
print('Start')
parser = argparse.ArgumentParser()
parser.add_argument('dump2_train1_test0', help='dump(2), train(1) or test(0)')
parser.add_argument('--model', help='model file')
parser.add_argument('--meta', help='metagraph file')
parser.add_argument('--output', help='output for dumping')
parser.add_argument('--gpu', help='the physical ids of GPUs to use')
parser.add_argument('--data', help='DAC dataset dir')
parser.add_argument('--run', help='directory of images to test')
parser.add_argument('--weights', help='weights file')
args = parser.parse_args()
print('Using GPU ' + str(args.gpu))
if args.gpu:
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
print(str(args.dump2_train1_test0))
if args.dump2_train1_test0 == '1':
if args.data == None:
print('Provide DAC dataset path with --data')
sys.exit()
config = get_config()
if args.model:
config.session_init = SaverRestore(args.model)
SimpleTrainer(config).train()
elif args.dump2_train1_test0 == '0':
if args.run == None:
print('Provide images with --run ')
sys.exit()
if args.weights == None:
print('Provide weights file (.npy) for testing!')
sys.exit()
assert args.weights.endswith('.npy')
run_image(Model(), DictRestore(np.load(args.weights, encoding='latin1').item()), args.run)
elif args.dump2_train1_test0 == '2':
if args.meta == None:
print('Provide meta file (.meta) for dumping')
sys.exit()
if args.model == None:
print('Provide model file (.data-00000-of-00001) for dumping')
sys.exit()
dump_weights(args.meta, args.model, args.output)
elif args.dump2_train1_test0 == '3':
if args.run == None:
print('Provide image with --run ')
sys.exit()
if args.weights == None:
print('Provide weights file (.npy) for testing!')
sys.exit()
assert args.weights.endswith('.npy')
run_single_image(Model(), DictRestore(np.load(args.weights, encoding='latin1').item()), args.run)
