Normalizing the keypoints/landmarks on an image to find its mean/std

[Mona_Jalal]

I am trying to normalize my targets (here landmark values) but to_tensor is not forcing the values to be between 0 and 1. How should that be enforced?

Also, am I looking for only one value for mean and one value for std, right? Each image has 4 landmarks and each landmark has x and y coordinates.

to_tensor = transforms.ToTensor()

landmarks_arr = []

for i in range(len(train_dataset)):
    landmarks = to_tensor(train_dataset[i]['landmarks'])
    landmarks[landmarks != landmarks] = 0  # this will set all nan to zero
    landmarks_arr.append(landmarks)
    
mean = torch.mean(torch.stack(landmarks_arr, dim=0))#, dim=(0, 2, 3))
std = torch.std(torch.stack(landmarks_arr, dim=0)) #, dim=(0, 2, 3))

print(mean.shape)
print("mean is {} and std is {}".format(mean, std))

result is:

torch.Size([])
mean is 367.7143527453419 and std is 160.8974583202625

This is how landmarks are:

landmarks_arr[0]

tensor([[[502.2869, 240.4949],
         [688.0000, 293.0000],
         [346.0000, 317.0000],
         [560.8283, 322.6830]]], dtype=torch.float64)

On a side note, is this line even correct?

mean = torch.mean(torch.stack(landmarks_arr, dim=0))#, dim=(0, 2, 3))

[Mona_Jalal]

Please let me know if this answer makes sense

landmarks_arr = []


for i in range(len(train_dataset)):
    landmarks = to_tensor(train_dataset[i]['landmarks'])
    landmarks[landmarks != landmarks] = 0  # this will set all nan to zero
    landmarks_arr.append(landmarks)
    
mean = torch.mean(torch.stack(landmarks_arr, dim=0))#, dim=(0, 2, 3))
std = torch.std(torch.stack(landmarks_arr, dim=0)) #, dim=(0, 2, 3))

max_val = 0.
tmp_max = 0.
for i in range(len(landmarks_arr)):
    tmp_max = torch.max(landmarks_arr[i])
    if max_val <= tmp_max:
        max_val = tmp_max
        
print('max val is: ', max_val)

print(mean.shape)
print("mean is {} and std is {}".format(mean/max_val, std/max_val))