def train_graph():
input_hyper = init_hyper()
input_pls = {
'feature':
tf.placeholder(dtype=tf.float32,
shape=[None, input_hyper['hidden_dim']])
}
vnet = VTranse()
vnet.create_graph(input_hyper['N_each_batch'], input_hyper['index_sp'],
input_hyper['index_cls'], input_hyper['N_cls'],
input_hyper['N_rela'])
graph_visual = Graph(input_hyper['layer_num'], input_hyper['hidden_dim'],
input_hyper['num_cls'], input_pls,
input_hyper['knowledge'])
graph_text = Graph(input_hyper['layer_num'], input_hyper['hidden_dim'],
input_hyper['num_cls'], input_pls,
input_hyper['knowledge'])
text_layer_out = graph_text.get_layer_out()
optimizer = tf.train.AdamOptimizer(learning_rate=input_hyper['lr_rate'])
train_var = tf.trainable_variables()
restore_var = [
var for var in train_var if 'vgg_16' in var.name or 'RD' in var.name
]
saver_res = tf.train.Saver(restore_var)
with tf.Session() as sess:
# init
init = tf.global_variables_initializer()
sess.run(init)
saver_res.restore(sess, input_hyper['model_path'])
roidb_read = read_roidb(input_hyper['roidb_path'])
train_roidb = roidb_read['train_roidb']
test_roidb = roidb_read['test_roidb']
N_train = len(train_roidb)
N_test = len(test_roidb)
for epoch in range(input_hyper['num_epoch']):
for roidb_id in range(N_train):
roidb_use = train_roidb[roidb_id]
if len(roidb_use['rela_gt']) == 0:
continue
rd_loss_temp, acc_temp, diff = vnet.train_predicate(
sess, roidb_use, None)
diff = np.array(diff)
print(diff.shape)
vf = []
print(np.array(input_hyper['rel_emb']).shape)
num_batch = diff.shape[0]
# for i in range(num_batch):
# num_nodes = diff[i][0]
#
# vf.append(visual_feature)
feed_dict = {}
feed_dict.update({input_pls['feature']: input_hyper['rel_emb']})
text_out = sess.run(text_layer_out, feed_dict=feed_dict)
print(text_out)
import sys
sys.path.append("models/ImgVRD/V-Trans-E")
from model.config import cfg
from model.ass_fun import *
from net.vtranse_vgg import VTranse
N_cls = cfg.VG_NUM_CLASS
N_rela = cfg.VG_NUM_RELA
N_each_batch = cfg.VG_BATCH_NUM_RELA
index_sp = False
index_cls = False
vnet = VTranse()
vnet.create_graph(N_each_batch, index_sp, index_cls, N_cls, N_rela)
roidb_path = cfg.DIR + 'vtranse/input/vg_rela_roidb.npz'
model_path = cfg.DIR + 'vtranse/pred_para/vg_vgg_rela/vg_vgg0001.ckpt'
save_path = cfg.DIR + 'vtranse/pred_res/vg_rela_roidb.npz'
roidb_read = read_roidb(roidb_path)
train_roidb = roidb_read['train_roidb']
test_roidb = roidb_read['test_roidb']
N_train = len(train_roidb)
N_test = len(test_roidb)
print('data loaded')
saver = tf.train.Saver()
from __future__ import print_function import tensorflow as tf import numpy as np from model.config import cfg from model.ass_fun import * from net.vtranse_vgg import VTranse N_cls = cfg.VRD_NUM_CLASS N_rela = cfg.VRD_NUM_RELA N_each_batch = cfg.VRD_BATCH_NUM_RELA lr_init = cfg.VRD_LR_INIT index_sp = False index_cls = True vnet = VTranse() vnet.create_graph(N_each_batch, index_sp, index_cls, N_cls, N_rela) roidb_path = cfg.DIR + 'vtranse/input/vrd_rela_roidb.npz' res_path = cfg.DIR + 'vtranse/pretrained_para/vrd_vgg_pretrained.ckpt' roidb_read = read_roidb(roidb_path) train_roidb = roidb_read['train_roidb'] test_roidb = roidb_read['test_roidb'] val_roidb = roidb_read['val_roidb'] N_train = len(train_roidb) N_val = len(val_roidb) N_round = cfg.VRD_TRAIN_ROUND N_show = 100 N_save = N_train
from __future__ import division from __future__ import print_function import tensorflow as tf import numpy as np from model.config import cfg from model.ass_fun import * from net.vtranse_vgg import VTranse N_cls = cfg.VRD_NUM_CLASS N_rela = cfg.VRD_NUM_RELA N_each_batch = cfg.VRD_BATCH_NUM index_sp = False index_cls = False vnet = VTranse() vnet.create_graph(N_each_batch, index_sp, index_cls, N_cls, N_rela) roidb_path = cfg.DIR + 'vtranse/input/vrd_roidb.npz' model_path = cfg.DIR + 'vtranse/pred_para/vrd_vgg/vrd_vgg0020.ckpt' save_path = cfg.DIR + 'vtranse/pred_res/vrd_pred_train_roidb.npz' roidb_read = read_roidb(roidb_path) # train_roidb = roidb_read['train_roidb'] test_roidb = roidb_read['train_roidb'] # N_train = len(train_roidb) N_test = len(test_roidb) saver = tf.train.Saver() with tf.Session() as sess:
