def export_outline_tan(folder, body_id, person_id):
args = {'folder': folder, 'person_id': person_id}
l_model = LinearModel(**args)
fbody = l_model.get_body(body_id, 'F')
sbody = l_model.get_body(body_id, 'S')
fbody.load_export_features()
sbody.load_export_features()
flag = 0
try:
base_path = os.path.join(config.tan_dir, folder)
if not os.path.exists(base_path):
os.mkdir(base_path)
outline_exp = OutlineTan(fbody, sbody)
file_path = os.path.join(base_path, '%s_FC.txt' % (body_id))
outline_exp.export_front(file_path)
file_path = os.path.join(base_path, '%s_SC.txt' % (body_id))
outline_exp.export_side(file_path)
except Exception as e:
logger.error('outline export fail: body_id=%s, person_id=%s' %
(body_id, person_id))
logger.error(e)
flag -= 1
try:
exporter = FeatureTan(fbody, sbody)
# sbody.features['f_neck_up_L'],_=exporter.map_front2side_feature('f_neck_up_L')
file_path = os.path.join(base_path, '%s_FL.txt' % (body_id))
exporter.write_file(file_path)
except Exception as e:
logger.error('feature export fail: body_id=%s, person_id=%s' %
(body_id, person_id))
logger.error(e)
flag -= 2
return flag
def get_front_side_points(name):
body_id = name
l_model = LinearModel()
fbody = l_model.get_body(body_id, 'F')
sbody = l_model.get_body(body_id, 'S')
fbody.load_export_features()
sbody.load_export_features()
outline_exp = OutlineTan(fbody, sbody)
return outline_exp.front_points(),outline_exp.side_points()
def feature_adjust_test(name):
body_id = name
l_model = LinearModel()
fbody = l_model.get_body(body_id, 'F')
sbody = l_model.get_body(body_id, 'S')
fbody.load_export_features()
sbody.load_export_features()
adjust = FeatureAdjust(fbody,sbody)
adjust.adjust()
def feature_test(name):
body_id = name
l_model = LinearModel()
fbody = l_model.get_body(body_id, 'F')
sbody = l_model.get_body(body_id, 'S')
fbody.load_export_features()
sbody.load_export_features()
exporter = FeatureTan(fbody, sbody)
features = exporter.export_features()
def outline_tan(folder, body_id, person_id):
global msg
args = {'folder': folder, 'person_id': person_id}
l_model = LinearModel(**args)
fbody = l_model.get_body(body_id, 'F')
sbody = l_model.get_body(body_id, 'S')
bbody = l_model.get_body(body_id, 'B')
b_outline_body = OutlineBody(bbody)
f_bd_features = fbody.bdfeatureXY
s_bd_features = sbody.bdfeatureXY
b_bd_features = bbody.bdfeatureXY
# fbody.load_export_features()
# sbody.load_export_features()
flag = 4
front = []
side = []
back = []
steps = 1
try:
outline_exp = OutlineTan(fbody, sbody)
front_points = outline_exp.front_points()
f_w = fbody.img_w
f_h = fbody.img_h
f_points = [dict(x=p[0], y=p[1]) for p in front_points]
front = dict(width=f_w, height=f_h, featureXY=f_points)
steps = 2
side_points = outline_exp.side_points()
s_w = sbody.img_w
s_h = sbody.img_h
s_points = [dict(x=p[0], y=p[1]) for p in side_points]
side = dict(width=s_w, height=s_h, featureXY=s_points)
steps = 3
back_points = b_outline_body.outline_points()
b_w = bbody.img_w
b_h = bbody.img_h
b_points = [dict(x=p[0], y=p[1]) for p in back_points]
back = dict(width=b_w, height=b_h, featureXY=b_points)
except Exception as e:
logger.error('outline export fail: body_id=%s, person_id=%s' %
(body_id, person_id))
logger.error(e)
flag = 5
msg = traceback.format_exc()
return flag, front, side, back, f_bd_features, s_bd_features, b_bd_features
def feature_tan(folder, body_id, person_id):
global msg
args = {'folder': folder, 'person_id': person_id}
l_model = LinearModel(**args)
fbody = l_model.get_body(body_id, 'F')
sbody = l_model.get_body(body_id, 'S')
fbody.load_export_features()
sbody.load_export_features()
flag = 4
features = {}
try:
exporter = FeatureTan(fbody, sbody)
features = exporter.export_features()
except Exception as e:
logger.error('feature export fail: body_id=%s, person_id=%s' %
(body_id, person_id))
logger.error(e)
flag = 5
msg = traceback.format_exc()
return flag, features
for seq in ind_seq:
ret.append(self.decode(seq))
return ret
else:
return self.ind2word[int(ind_seq)]
if __name__ == '__main__':
args = parseArgs()
# load vocab data
with open('./data/VocabData.pkl', 'rb') as f:
VocabData = pickle.load(f)
# load linear model, transform feature tensor to semantic space
linNet = LinearModel(hiddenSize=4096)
sos_id = VocabData['word_dict']['<START>']
eos_id = VocabData['word_dict']['<END>']
lstmDec = DecoderRNN(vocab_size=len(VocabData['word_dict']),max_len=15,sos_id=sos_id, eos_id=eos_id , embedding_size=300,hidden_size=4096,
embedding_parameter=VocabData['word_embs'], update_embedding=False ,use_attention=True)
# todo: reload lstmEnc
linNet, lstmDec = reloadModel(args.model_path, linNet, lstmDec)
loader = LoaderDemo()
# loader = DataLoader(dataset, batch_size=args.batch_imgs, shuffle=False, num_workers=2,
# collate_fn=dataset.collate_fn)
symbolDec = SymbolDecoder(VocabData['word_dict'])
default='./save/default/')
parser.add_argument('-b','--batch_imgs',
default=4, type=int)
args = parser.parse_args()
return args
if __name__ == '__main__':
args = parseArgs()
# load vocab data
with open('./data/VocabData.pkl', 'rb') as f:
VocabData = pickle.load(f)
# load linear model, transform feature tensor to semantic space
linNet = LinearModel(hiddenSize=4096)
# load LSTM encoder
lstmEnc = EncoderRNN(len(VocabData['word_dict']), 15, 4096, 300,
input_dropout_p=0, dropout_p=0,
n_layers=1, bidirectional=False, rnn_cell='lstm', variable_lengths=True,
embedding_parameter=VocabData['word_embs'], update_embedding=False)
# load crit
crit = SimilarityLoss(0.5,0.5,1)
if args.evaluate_mode: # evaluation mode
loader = LoaderEnc(mode='test')
linNet,lstmEnc = reloadModel(args.model_path,linNet,lstmEnc)
eval(loader,linNet,lstmEnc,crit)
else: # train mode
optimizer = torch.optim.Adam(list(filter(lambda p: p.requires_grad, lstmEnc.parameters()))+list(linNet.parameters()), 0.0001)
dataset = LoaderEnc()
parser.add_argument('-s', '--save_path', default='./save/default/')
parser.add_argument('-b', '--batch_imgs', default=4, type=int)
args = parser.parse_args()
return args
if __name__ == '__main__':
args = parseArgs()
# load vocab data
with open('./data/VocabData.pkl', 'rb') as f:
VocabData = pickle.load(f)
# load linear model, transform feature tensor to semantic space
linNet = LinearModel(hiddenSize=4096)
# load LSTM encoder
lstmEnc = EncoderRNN(len(VocabData['word_dict']),
15,
4096,
300,
input_dropout_p=0,
dropout_p=0,
n_layers=1,
bidirectional=False,
rnn_cell='lstm',
variable_lengths=True,
embedding_parameter=VocabData['word_embs'],
update_embedding=False)
# load crit
crit = SimilarityLoss(0.5, 0.5, 1)
parser.add_argument('-c', '--cont_model_path',
default='./save/default/lstmDec.pt')
args = parser.parse_args()
return args
if __name__ == '__main__':
args = parseArgs()
# load vocab data
with open('./data/VocabData.pkl', 'rb') as f:
VocabData = pickle.load(f)
# load linear model, transform feature tensor to semantic space
linNet = LinearModel()
sos_id = VocabData['word_dict']['<START>']
eos_id = VocabData['word_dict']['<END>']
# load LSTM encoder
lstmDec = DecoderRNN(vocab_size=len(VocabData['word_dict']),max_len=15,sos_id=sos_id, eos_id=eos_id , embedding_size=300,hidden_size=1024,
embedding_parameter=VocabData['word_embs'], update_embedding=False ,use_attention=True,use_prob_vector=True)
"""
lstmEnc = EncoderRNN(len(VocabData['word_dict']), max_len=15, hidden_size=4096, embedding_size=300,
input_dropout_p=0, dropout_p=0,
n_layers=1, bidirectional=False, rnn_cell='lstm', variable_lengths=False,
embedding_parameter=VocabData['word_embs'], update_embedding=False)
# todo: reload lstmEnc
#linNet,lstmEnc = reloadModel(args.model_path, linNet, lstmEnc)
"""