def __init__(self, words_list):
super(Net, self).__init__()
question_features = 1024
vision_features = config.output_features
glimpses = 2
self.text = word_embedding.TextProcessor(
classes=words_list,
embedding_features=300,
lstm_features=question_features,
drop=0.0,
)
self.attention = Attention(
v_features=vision_features,
q_features=question_features,
mid_features=1024,
glimpses=glimpses,
drop=0.2,
)
self.classifier = Classifier(
in_features=(glimpses * vision_features, question_features),
mid_features=1024,
out_features=config.max_answers,
drop=0.5,
)
def __init__(self, words_list):
super(Net, self).__init__()
self.question_features = config.question_features
self.vision_features = config.output_features
self.spatial_features = config.spatial_features
self.hidden_features = config.hidden_features
self.num_inter_head = config.num_inter_head
self.num_intra_head = config.num_intra_head
self.num_block = config.num_block
self.spa_block = config.spa_block
self.que_block = config.que_block
self.visual_normalization = config.visual_normalization
self.iteration = config.iteration
assert (self.hidden_features % self.num_inter_head == 0)
assert (self.hidden_features % self.num_intra_head == 0)
words_list = list(words_list)
words_list.insert(
0, '__unknown__') # add 'unk' key to the available vocab list
self.text = word_embedding.TextProcessor(
classes=words_list,
embedding_features=300,
lstm_features=self.question_features,
drop=0.1,
)
self.interIntraBlocks = SingleBlock(
num_block=self.num_block, # 2
spa_block=self.spa_block, # 2
que_block=self.que_block, # 2
iteration=self.iteration, # 2
v_size=self.vision_features, # 2048
q_size=self.question_features, # 1280
b_size=self.spatial_features, # 4
output_size=self.hidden_features, # 512
num_inter_head=self.num_inter_head, # 8
num_intra_head=self.num_intra_head, # 8
drop=0.1,
)
self.classifier = Classifier(in_features=self.hidden_features,
mid_features=config.mid_features,
out_features=config.max_answers,
drop=config.classifier_dropout)
for m in self.modules():
if isinstance(m, nn.Linear) or isinstance(m, nn.Conv2d):
init.xavier_uniform_(m.weight)
if m.bias is not None:
m.bias.data.zero_()
def __init__(self, words_list):
super(Net, self).__init__()
mid_features = 1024
question_features = mid_features
vision_features = config.output_features
self.top_k_sparse = 16
num_kernels = 8
sparse_graph = True
self.text = word_embedding.TextProcessor(
classes=words_list,
embedding_features=300,
lstm_features=question_features,
drop=0.0,
)
self.pseudo_coord = PseudoCoord()
self.graph_learner = GraphLearner(
v_features=vision_features + 4,
q_features=question_features,
mid_features=512,
dropout=0.5,
sparse_graph=sparse_graph,
)
self.graph_conv1 = GraphConv(v_features=vision_features + 4,
mid_features=mid_features * 2,
num_kernels=num_kernels,
bias=False)
self.graph_conv2 = GraphConv(v_features=mid_features * 2,
mid_features=mid_features,
num_kernels=num_kernels,
bias=False)
self.classifier = Classifier(
in_features=mid_features,
mid_features=mid_features * 2,
out_features=config.max_answers,
drop=0.5,
)
self.relu = nn.ReLU()
self.dropout = nn.Dropout(0.5)
def __init__(self, words_list):
super(Net, self).__init__()
self.question_features = 1280
self.vision_features = config.output_features
self.hidden_features = 512
self.num_inter_head = 8
self.num_intra_head = 8
self.num_block = 2
self.visual_normalization = True
assert(self.hidden_features % self.num_inter_head == 0)
assert(self.hidden_features % self.num_intra_head == 0)
words_list = list(words_list)
words_list.insert(0, '__unknown__')
self.text = word_embedding.TextProcessor(
classes=words_list,
embedding_features=300,
lstm_features=self.question_features,
drop=0.1,
)
self.interIntraBlocks = SingleBlock(
num_block=self.num_block,
v_size=self.vision_features,
q_size=self.question_features,
output_size=self.hidden_features,
num_inter_head=self.num_inter_head,
num_intra_head=self.num_intra_head,
drop=0.1,
)
self.classifier = Classifier(
in_features=self.hidden_features,
mid_features=1024,
out_features=config.max_answers,
drop=0.1,)
for m in self.modules():
if isinstance(m, nn.Linear) or isinstance(m, nn.Conv2d):
init.xavier_uniform_(m.weight)
if m.bias is not None:
m.bias.data.zero_()
def __init__(self, words_list):
super(Net, self).__init__()
num_hid = 1280
question_features = num_hid
vision_features = config.output_features
glimpses = 12
objects = 10
self.text = word_embedding.TextProcessor(
classes=words_list,
embedding_features=300,
lstm_features=question_features,
use_hidden=False,
drop=0.0,
)
self.count = Counter(objects)
self.attention = BiAttention(
v_features=vision_features,
q_features=question_features,
mid_features=num_hid,
glimpses=glimpses,
drop=0.5,)
self.apply_attention = ApplyAttention(
v_features=vision_features,
q_features=question_features,
mid_features=num_hid,
glimpses=glimpses,
num_obj=objects,
count = self.count,
drop=0.2,
)
self.classifier = Classifier(
in_features=num_hid,
mid_features=num_hid * 2,
out_features=config.max_answers,
drop=0.5,)
def __init__(self, words_list):
super(Net, self).__init__()
self.question_features = 1280
self.vision_features = config.output_features
self.hidden_features = 512
self.num_inter_head = 8
self.num_intra_head = 8
self.num_block = 1
assert (self.hidden_features % self.num_inter_head == 0)
assert (self.hidden_features % self.num_intra_head == 0)
self.text = word_embedding.TextProcessor(
classes=words_list,
embedding_features=300,
lstm_features=self.question_features,
use_hidden=False, # use whole output, not just final hidden
drop=0.0,
)
self.interIntraBlocks = MultiBlock(
num_block=self.num_block,
v_size=self.vision_features,
q_size=self.question_features,
output_size=self.hidden_features,
num_inter_head=self.num_inter_head,
num_intra_head=self.num_intra_head,
drop=0.1,
)
self.classifier = Classifier(
in_features=self.hidden_features,
mid_features=2048,
out_features=config.max_answers,
drop=0.5,
)