-
Notifications
You must be signed in to change notification settings - Fork 21
/
model.py
200 lines (166 loc) · 8.81 KB
/
model.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
from __future__ import division
import tensorflow as tf
import numpy as np
import cPickle as pickle
import os, sys
import scipy.io
import time
from util.rnn import lstm_layer as lstm
from util.rnn import bi_lstm_layer as bi_lstm
from util.cnn import fc_relu_layer as fc_relu
from util.cnn import fc_layer as fc
from util.cnn import conv_layer as conv
from util.bn import batch_norm as bn
from util.custom_init import msr_init
from util import loss as loss_func
class ground_model(object):
def __init__(self, config=None):
self.batch_size = self._init_param(config, 'batch_size', 20)
self.test_batch_size = self._init_param(config, 'test_batch_size', -1)
self.class_num = self._init_param(config, 'class_num', 100)
self.lr = self._init_param(config, 'lr', 0.0001)
self.init = self._init_param(config, 'init', 'axvier')
self.optim = self._init_param(config, 'optim', 'adam')
self.vocab_size = self._init_param(config, 'vocab_size', 17150)
self.img_feat_size = self._init_param(config, 'img_feat_size', 4096+5) # bbx info included in visual feature
self.dropout = self._init_param(config, 'dropout', 0.5)
self.num_lstm_layer = self._init_param(config, 'num_lstm_layer', 1)
self.num_prop = self._init_param(config, 'num_prop', 100)
self.lstm_dim = self._init_param(config, 'lstm_dim', 500)
self.hidden_size = self._init_param(config, 'hidden_size', 128)
self.phrase_len = self._init_param(config, 'phrase_len', 19)
self.weight_decay = self._init_param(config, 'weight_decay', 0.0005)
self.reg_lambda = self._init_param(config, 'reg_lambda', 1.0)
self.embed_w = self._init_param(config, 'embed_w', None)
self.top_k = self._init_param(config, 'top_k', 10)
self.is_multi = self._init_param(config, 'is_multi', True)
def _init_param(self, config, param_name, default_value):
if hasattr(config, param_name):
return getattr(config, param_name)
else:
return default_value
def init_placeholder(self):
self.sen_data = tf.placeholder(tf.int32, [self.batch_size, self.phrase_len])
self.vis_data = tf.placeholder(tf.float32, [self.batch_size, self.num_prop, self.img_feat_size])
self.bbx_label = tf.placeholder(tf.int32, [self.batch_size])
self.gt_reg = tf.placeholder(tf.float32, [self.batch_size, 4])
self.is_train = tf.placeholder(tf.bool)
self.pos_all = tf.placeholder(tf.int32, [None, 2])
self.pos_reg_all = tf.placeholder(tf.float32, [None, 4])
self.num_reg = tf.placeholder(tf.float32)
self.reward = tf.placeholder(tf.float32, [None, self.num_prop])
def model_structure(self, sen_data, vis_data, batch_size, is_train, dropout=None):
if dropout == None:
dropout = self.dropout
text_seq_batch = tf.transpose(sen_data, [1, 0]) # input data is [num_steps, batch_size]
with tf.variable_scope('word_embedding'), tf.device("/cpu:0"):
if self.embed_w is None:
initializer = tf.contrib.layers.xavier_initializer(uniform=True)
else:
initializer = tf.constant_initializer(self.embed_w)
embedding_mat = tf.get_variable("embedding", [self.vocab_size, self.lstm_dim], tf.float32,
initializer=initializer)
# text_seq has shape [T, N] and embedded_seq has shape [T, N, D].
embedded_seq = tf.nn.embedding_lookup(embedding_mat, text_seq_batch)
# encode phrase based on the last step of hidden states
outputs, _, _ = bi_lstm('lstm_lang', embedded_seq, None, output_dim=self.lstm_dim,
num_layers=1, forget_bias=1.0, apply_dropout=False,concat_output=False,
initializer=tf.random_uniform_initializer(minval=-0.08, maxval=0.08))
sen_raw = outputs[-1]
vis_raw = tf.reshape(vis_data, [self.batch_size*self.num_prop, self.img_feat_size])
sen_bn = bn(sen_raw, is_train, "SEN_BN", 0.9)
vis_bn = bn(vis_raw, is_train, "VIS_BN", 0.9)
sen_output = tf.reshape(sen_bn, [self.batch_size, 1, 1, 2*self.lstm_dim]) # bi-directional lstm: hidden_size double
vis_output = tf.reshape(vis_bn, [self.batch_size, self.num_prop, 1, self.img_feat_size])
sen_tile = tf.tile(sen_output, [1, self.num_prop, 1, 1])
feat_concat = tf.concat(3, [sen_tile, vis_output])
feat_proj_init = msr_init([1, 1, 2*self.lstm_dim+self.img_feat_size, self.hidden_size])
feat_proj = conv("feat_proj", feat_concat, 1, 1, self.hidden_size, weights_initializer=feat_proj_init)
feat_relu = tf.nn.relu(feat_proj)
att_conv_init = msr_init([1, 1, self.hidden_size, 5])
att_conv = conv("att_conv", feat_relu, 1, 1, 5, weights_initializer=att_conv_init)
att_scores = tf.reshape(att_conv, [self.batch_size, self.num_prop, 5])
att_logits = tf.reshape(att_scores[:, :, 0], [self.batch_size, self.num_prop])
_, pred_ind = tf.nn.top_k(att_logits, self.top_k)
pred_ind = tf.reshape(pred_ind, [self.batch_size*self.top_k, 1])
row_ind = tf.reshape(tf.range(0, self.batch_size), [-1, 1])
row_ind = tf.reshape(tf.tile(row_ind, [1, self.top_k]), [self.top_k*self.batch_size, 1])
pred_ind = tf.concat(1, [row_ind, pred_ind])
# (batch_size*top_k) x img_feat_size
vis_top = tf.gather_nd(tf.reshape(vis_output, [self.batch_size, self.num_prop, self.img_feat_size]), pred_ind)
vis_ref = tf.reduce_mean(tf.reshape(vis_top, [self.batch_size, self.top_k, self.img_feat_size]), 1)
ref_feat = tf.concat(1, [vis_ref, sen_bn])
# ref_feat = vis_ref
reward_pred = tf.reshape(tf.sigmoid(fc('reward_pred', ref_feat, 1)),[self.batch_size])
return att_scores, reward_pred
def train_loss(self, att_scores, labels, reward, reward_pred, is_multi=False, pos_all=None, pos_reg_all=None, num_reg=None):
att_logits = tf.reshape(att_scores[:, :, 0], [self.batch_size, self.num_prop])
loss_vec=tf.nn.sparse_softmax_cross_entropy_with_logits(att_logits, labels, name=None)
loss_cls = tf.reduce_mean(loss_vec)
# calculate reward for top k predictions for reinforcement learning
_, pred_ind = tf.nn.top_k(att_logits, self.top_k)
pred_ind = tf.reshape(pred_ind, [self.batch_size*self.top_k, 1])
row_ind = tf.reshape(tf.range(0, self.batch_size), [-1, 1])
row_ind = tf.reshape(tf.tile(row_ind, [1, self.top_k]), [self.top_k*self.batch_size, 1])
pred_ind = tf.concat(1, [row_ind, pred_ind])
pred_reward = tf.gather_nd(reward, pred_ind)
pred_reward = tf.reshape(pred_reward, [self.batch_size, self.top_k])
reward_weight = tf.reduce_mean(pred_reward, 1)
if is_multi:
pred_reg = tf.gather_nd(att_scores[:, :, 1:], pos_all)
loss_reg = loss_func.smooth_l1_regression_loss(pred_reg, pos_reg_all)/num_reg
else:
pred_label = tf.cast(tf.reshape(tf.argmax(att_logits, 1), [-1, 1]), tf.int32)
# pred_label = labels
row_index = tf.reshape(tf.range(0, self.batch_size), [-1, 1])
pred_index = tf.concat(1, [row_index, pred_label])
pred_reg = tf.gather_nd(att_scores[:, :, 1:], pred_index)
loss_reg = loss_func.smooth_l1_regression_loss(pred_reg, self.gt_reg)
loss = loss_cls + self.reg_lambda*loss_reg
loss_rwd = tf.nn.l2_loss(reward_weight-reward_pred)
return loss, loss_vec, reward_weight, loss_rwd
def get_variables_by_name(self,name_list, verbose=True):
v_list=tf.trainable_variables()
v_dict={}
for name in name_list:
v_dict[name]=[]
for v in v_list:
for name in name_list:
if name in v.name: v_dict[name].append(v)
#print
if verbose:
for name in name_list:
print "Variables of <"+name+">"
for v in v_dict[name]:
print " "+v.name
return v_dict
def build_train_op(self, loss, loss_vec, loss_rwd, reward_pred):
if self.optim == 'adam':
print 'Adam optimizer'
v_dict = self.get_variables_by_name([""])
tvars = [v for v in v_dict[""] if "reward" not in v.name]
optim_all = tf.train.AdamOptimizer(self.lr,name='Adam_all')
grads_all = tf.gradients(loss, tvars)
train_op_all = optim_all.apply_gradients(zip(grads_all, tvars))
optim_ref = tf.train.AdamOptimizer(self.lr*0.01,name='SGD_ref')
att_vars = self.get_variables_by_name(["att"])["att"]
grad_ref = tf.gradients(loss, att_vars, grad_ys=tf.reduce_mean(reward_pred))
train_op_ref = optim_ref.apply_gradients(zip(grad_ref, att_vars))
optim_rwd = tf.train.AdamOptimizer(self.lr,name="Adam_rwd")
rwd_vars = self.get_variables_by_name(["reward"])["reward"]
grad_rwd = tf.gradients(loss_rwd, rwd_vars)
train_op_rwd = optim_rwd.apply_gradients(zip(grad_rwd, rwd_vars))
train_op = tf.group(train_op_all, train_op_ref, train_op_rwd)
else:
print 'SGD optimizer'
tvars = tf.trainable_variables()
optimizer = tf.train.GradientDescentOptimizer(self._lr)
grads = tf.gradients(loss, tvars)
train_op = optimizer.apply_gradients(zip(grads, tvars))
return train_op
def build_model(self):
self.init_placeholder()
att_logits, reward_pred = self.model_structure(self.sen_data, self.vis_data, self.batch_size, self.is_train)
self.loss, loss_vec, self.reward_w, loss_rwd = self.train_loss(att_logits, self.bbx_label, self.reward, reward_pred, self.is_multi, self.pos_all, self.pos_reg_all, self.num_reg)
# self.train_op = self.build_train_op(self.loss, loss_vec, self.reward_w)
return self.loss, loss_vec, att_logits, reward_pred, loss_rwd