def forward(self, x1, finger, x3):
protein = x3.view(x3.size(0), -1).to(device)
x1 = torch.unsqueeze(x1, 1)
pic = self.conv(x1)
# print(x1.shape, pic.shape)
pic = self.bn(pic)
pic = self.r_elu(pic)
pic = self.layer1(pic)
pic = self.layer2(pic)
spp = spatial_pyramid_pool(
pic, pic.size(0),
[int(pic.size(2)), int(pic.size(3))], self.output_num)
# print(spp.shape, "spp.shape")
# print(spp.shape, "spp.shape")
fc1 = F.relu(self.fc1(spp))
fc2 = F.relu(self.fc2(fc1))
sscomplex = torch.cat([fc2, finger, protein], dim=1)
sscomplex = torch.relu(self.linear_final_step(sscomplex))
if not bool(self.type):
pred = self.linear_final(sscomplex)
pic_output = torch.sigmoid(pred)
return pic_output
def __init__(self,
input_nc=1,
ndf=64,
gpu_ids=[],
num_class=2): #initialized variables
super(Network, self).__init__()
self.gpu_ids = gpu_ids
self.output_num = (4, 2, 2, 1)
# Use FIVE conv layers (conv, relu, maxpool)
self.conv_layer1 = self._make_conv_layer(3, 32)
self.conv_layer2 = self._make_conv_layer(32, 64)
self.conv_layer3 = self._make_conv_layer(64, 124)
self.conv_layer4 = self._make_conv_layer(124, 256)
# Use FIVE max pooling layers
# One 3D SPP layer
self.spp_layer = spatial_pyramid_pool(self=None,
previous_conv=x,
num_sample=1,
previous_conv_size=[
int(x.size(2)),
int(x.size(3)),
int(x.size(4))
],
out_pool_size=self.output_num)
# Two Fully Connected Layers
self.fc1 = nn.Linear(4096, 2048)
self.fc2 = nn.Linear(1024, 512)
self.fc2 = nn.Linear(256, num_class)
def through_spp(self, x): #spp_layer
for i in range(BATCH_SIZE):
y_piece = torch.unsqueeze(spatial_pyramid_pool(previous_conv = x[i,:], num_sample = R,
previous_conv_size = [x.size(3),x.size(4)], out_pool_size = [2, 2]), 0)
if i == 0:
y = y_piece
#print(y_piece.shape)
else:
y = torch.cat((y, y_piece))
#print(y.shape)
return y
def forward(self, x):
x = self.conv1(x)
output_num = [8, 4, 1]
x_spp = spatial_pyramid_pool(
x, self.batch_size,
[int(x.size(2)), int(x.size(3))], output_num)
x = self.liner(x_spp)
x = self.out(x)
return x
def forward(self, x):
out = self.conv1(x)
out = self.conv2(out)
# Flatten
# print(out.view(out.size(0), -1).size())
# print(out.size())
feature = spatial_pyramid_pool(
self, out, out.size(0),
[int(out.size(2)), int(out.size(3))], self.output_num)
out = self.fc1(feature)
out = self.fc2(out)
return out, feature
def forward(self, x1, x2):
x1 = torch.unsqueeze(x1, 1)
drug1 = self.conv_drug1(x1)
# print(x1.shape, pic.shape)
drug1 = self.bn_drug1(drug1)
drug1 = self.elu_drug1(drug1)
drug1 = self.layer1_drug1(drug1)
drug1 = self.layer2_drug1(drug1)
drug2 = torch.unsqueeze(x2, 1)
drug2 = self.conv_drug2(drug2)
# print(x1.shape, pic.shape)
drug2 = self.bn_drug2(drug2)
drug2 = self.elu_drug2(drug2)
drug2 = self.layer1_drug2(drug2)
drug2 = self.layer2_drug2(drug2)
# print(pic.shape, "pic.shape")
spp_drug1 = spatial_pyramid_pool(drug1, drug1.size(0), [int(drug1.size(2)), int(drug1.size(3))],
self.output_num)
spp_drug2 = spatial_pyramid_pool(drug2, drug2.size(0), [int(drug2.size(2)), int(drug2.size(3))],
self.output_num)
# print(spp.shape, "spp.shape")
fc1_drug1 = F.relu(self.fc1(spp_drug1))
fc2_drug1 = F.relu(self.fc2(fc1_drug1))
fc1_drug2 = F.relu(self.fc1(spp_drug2))
fc2_drug2 = F.relu(self.fc2(fc1_drug2))
# print(fc2.shape, "fc2.shape")
sscomplex = torch.cat([fc2_drug1, fc2_drug2], dim=1)
sscomplex = torch.relu(self.linear_final_step(sscomplex))
if not bool(self.type):
pred = self.linear_final(sscomplex)
pic_output = torch.sigmoid(pred)
return pic_output
def through_spp_new(self, x, ssw): #x.shape = [BATCH_SIZE, 512, 14, 14] ssw_get.shape = [BATCH_SIZE, R, 4] y.shape = [BATCH_SIZE, R, 4096]
for i in range(BATCH_SIZE):
for j in range(ssw.size(1)):
fmap_piece = torch.unsqueeze(x[i, :, floor(ssw[i, j, 0]) : floor(ssw[i, j, 0] + ssw[i, j, 2]),
floor(ssw[i, j, 1]) : floor(ssw[i, j, 1] + ssw[i, j, 3])], 0)
fmap_piece = spatial_pyramid_pool(previous_conv = fmap_piece, num_sample = 1,
previous_conv_size = [fmap_piece.size(2),fmap_piece.size(3)], out_pool_size = [2, 2])
if j == 0:
y_piece = fmap_piece
#print('fmap_piece.shape', fmap_piece.shape)
else:
y_piece = torch.cat((y_piece, fmap_piece))
if i == 0:
y = torch.unsqueeze(y_piece, 0)
#print('y_piece', y_piece.shape)
else:
y = torch.cat((y, torch.unsqueeze(y_piece, 0)))
return y
def forward(self, x):
x = self.conv1(x)
x = self.LReLU1(x)
x = self.conv2(x)
x = F.leaky_relu(self.BN1(x))
x = self.conv3(x)
x = F.leaky_relu(self.BN2(x))
x = self.conv4(x)
# x = F.leaky_relu(self.BN3(x))
# x = self.conv5(x)
spp = spatial_pyramid_pool(
x, 1, [int(x.size(2)), int(x.size(3))], self.output_num)
# print(spp.size())
fc1 = self.fc1(spp)
fc2 = self.fc2(fc1)
s = nn.Sigmoid()
output = s(fc2)
return output