fg91

class YourCustomModel(nn.Module):
    def __init__(self):
     	super().__init__()
	# truncated base network, „True“ refers to pretrained
    	self.backbone = nn.Sequential(*list(resnet34(True).children())[:8])
     
	# and your custom layers
	self.features = nn.Sequential(
   	self.backbone,
     	# custom layers:

fg91

class SaveFeatures():
    def __init__(self, module):
        self.hook = module.register_forward_hook(self.hook_fn)
    def hook_fn(self, module, input, output):
        self.features = torch.tensor(output,requires_grad=True).cuda()
    def close(self):
        self.hook.remove()

fg91

class FilterVisualizer():
    def __init__(self, size=56, upscaling_steps=12, upscaling_factor=1.2):
        self.size, self.upscaling_steps, self.upscaling_factor = size, upscaling_steps, upscaling_factor
        self.model = vgg16(pre=True).cuda().eval()
        set_trainable(self.model, False)

    def visualize(self, layer, filter, lr=0.1, opt_steps=20, blur=None):
        sz = self.size
        img = np.uint8(np.random.uniform(150, 180, (sz, sz, 3)))/255  # generate random image
        activations = SaveFeatures(list(self.model.children())[layer])  # register hook

fg91

layer = 40
filter = 265
FV = FilterVisualizer(size=56, upscaling_steps=12, upscaling_factor=1.2)
FV.visualize(layer, filter, blur=5)

fg91

model = models.resnet34(pretrained=True).cuda()

# Freeze the base of the network and only train the new custom layers
for param in model.parameters():
    param.requires_grad = False
    
p=0.1

model.fc = nn.Sequential(nn.BatchNorm1d(512),
                         nn.Dropout(p),

fg91

def cos_loss(input, target):
    return 1 - F.cosine_similarity(input, target).mean()

fg91

from x import task2


@task
def my_task():
  ...
  
 
def relocate(task, context="."):