‘’’
class Orfunction(torch.nn.Module):
def __init__(self, D_in, H, num_classes):
super().__init__()
self.D_in = D_in
self.H = H
self.D_out = num_classes
self.conv1 = nn.Conv2d(D_in,H,kernel_size=11, stride=4, padding=2)
self.mpool1 = nn.MaxPool2d(kernel_size=3, stride=2)
self.conv2 = nn.Conv2d(64, 192, kernel_size=5, padding=2)
self.mpool2 = nn.MaxPool2d(kernel_size=3, stride=2)
self.conv3 = nn.Conv2d(192, 384, kernel_size=3, padding=1)
self.conv4 = nn.Conv2d(384, 256, kernel_size=3, padding=1)
self.conv5 = nn.Conv2d(256, 256, kernel_size=3, padding=1)
self.mpool3 = nn.MaxPool2d(kernel_size=3, stride=2)
self.avgpool = nn.AdaptiveAvgPool2d((6, 6))
self.linear1 = nn.Linear(256 * 6 * 6, 4096)
self.linear2 = nn.Linear(4096, 4096)
self.linear3 = nn.Linear(4096, num_classes)
def forward(self, x):
#print(self.conv1.weight)
x = self.conv1(x)
x = F.leaky_relu(x)
x = self.mpool1(x)
x = self.conv2(x)
x = F.leaky_relu(x)
x = self.mpool2(x)
x = self.conv3(x)
x = F.leaky_relu(x)
x = self.conv4(x)
x = F.leaky_relu(x)
x = self.conv5(x)
x = F.leaky_relu(x)
x = self.mpool3(x)
x = self.avgpool(x)
x = x.view(x.size(0), 256 * 6 * 6)
x = self.linear1(x)
x = F.leaky_relu(x)
x = self.linear2(x)
x = F.leaky_relu(x)
x = self.linear3(x)
x = torch.softmax(x,1)
#print("s",x.shape)
#print("pobability distribution is:",x)
return x
D_in = 3
H = 64
num_classes = 3
classes = [‘truck’,‘auto’,‘car’]
function_network = Orfunction(D_in,H,num_classes)
Device = torch.device(‘cpu’)
function_network.to(Device)
loss_function = torch.nn.CrossEntropyLoss()
vls = list(function_network.parameters())
optimizer = optim.Adam(vls, lr=0.01, weight_decay=0.001)
list1 = list()
path2 = ‘/home/prashant/racing_car/google-images-download/downloads/train’
images = os.listdir(path2)
for image in images:
img = os.path.join(path2,image)
list1.append(img)
list2 = list()
for imgs in list1:
img = cv2.imread(imgs)
img1 = cv2.resize(img, (256, 256))
list2.append(img1)
list3 = list()
for j in list2:
img1 = torch.tensor(j, dtype= torch.float,requires_grad=True)
img1 = img1.permute(2,0,1)
img1 = img1.unsqueeze(dim = 0)
list3.append(img1)
batch1 = torch.stack(list3,dim = 1)
batch1 = batch1.squeeze(dim = 0)
y = torch.tensor([0,0,0,0,0,1,1,1,1,1,2,2,2,2,2])
print(y.shape)
epochs = 20
for i in range(epochs):
print('iteration', i)
optimizer.zero_grad()
output = function_network(batch1)
print("o",output.shape)
loss = loss_function(output,y)
loss = Variable(loss,requires_grad=True)
print("loss", loss.item())
print(list(function_network.parameters())[0].grad)
init1 = list(function_network.parameters())[0].clone()
loss.backward()
optimizer.step()
init2 = list(function_network.parameters())[0].clone()
t = torch.equal(init1.data, init2.data)
print('are the same :',t)
"output is :
loss 1.099882960319519 "
are they same : True
loss 1.099882960319519 "
are they same : True
loss 1.099882960319519 "
are they same : True
loss 1.099882960319519 "
are they same : True
loss 1.099882960319519 "
are they same : True
loss 1.099882960319519 "
are they same : True
loss 1.099882960319519 "
are they same : True
loss 1.099882960319519 "
are they same : True
‘’’