Untitled

# This obtains the list of known identities from the known folder
known_regex = "/content/data/11-785-s23-hw2p2-verification/known/*/*"
known_paths = [i.split('/')[-2] for i in sorted(glob.glob(known_regex))]

# Obtain a list of images from unknown folders
unknown_dev_regex = "/content/data/11-785-s23-hw2p2-verification/unknown_dev/*"
unknown_test_regex = "/content/data/11-785-s23-hw2p2-verification/unknown_test/*"

# We load the images from known and unknown folders
unknown_dev_images = [Image.open(p) for p in tqdm(sorted(glob.glob(unknown_dev_regex)))]
unknown_test_images = [Image.open(p) for p in tqdm(sorted(glob.glob(unknown_test_regex)))]
known_images = [Image.open(p) for p in tqdm(sorted(glob.glob(known_regex)))]

# Why do you need only ToTensor() here?
transforms = torchvision.transforms.Compose([
    torchvision.transforms.ToTensor(),
    normalize])

unknown_dev_images = torch.stack([transforms(x) for x in unknown_dev_images])
unknown_test_images = torch.stack([transforms(x) for x in unknown_test_images])
known_images  = torch.stack([transforms(y) for y in known_images ])
#Print your shapes here to understand what we have done

# You can use other similarity metrics like Euclidean Distance if you wish
similarity_metric = torch.nn.CosineSimilarity(dim= 1, eps= 1e-6) 
# similarity_metric = torch.nn.PairwiseDistance()