import os
import urllib.request

if not os.path.exists("data"):
    os.mkdir("data")
    urllib.request.urlretrieve("https://upload.wikimedia.org/wikipedia/commons/b/bb/Andy_Lau_%E5%88%98%E5%BE%B7%E5%8D%8E%2C_Beijing_International_Film_Festival_%E5%8C%97%E4%BA%AC%E7%94%B5%E5%BD%B1%E8%8A%82%2C_2013_%28cropped%29.jpg", "data/劉德華.jpg")
    urllib.request.urlretrieve("https://upload.wikimedia.org/wikipedia/commons/d/dc/Chaplin_The_Champion.jpg", "data/卓別林.jpg")
    urllib.request.urlretrieve("https://upload.wikimedia.org/wikipedia/commons/thumb/8/86/Dayo_Wong_at_Olympian_City.jpg/800px-Dayo_Wong_at_Olympian_City.jpg", "data/黃子華.jpg")
    urllib.request.urlretrieve("https://upload.wikimedia.org/wikipedia/commons/2/2c/Woody_Allen_Cannes_2015.jpg", "data/伍迪·艾伦.jpg")
    urllib.request.urlretrieve("https://upload.wikimedia.org/wikipedia/commons/0/09/RussellPeters08TIFF.jpg", "data/罗素·彼得斯.jpg")
    opener = urllib.request.URLopener()
    opener.addheader('User-Agent', 'whatever')
    opener.retrieve("https://cfcdn.she.com/media/she.com/2023/03/411903-411903-Screenshot-2023-03-10-at-11.19.32-940x1024.png", "data/林青霞.png")
    opener.retrieve("https://cfcdn.she.com/media/she.com/2023/03/122403-122403-IMG_4831-481x600-1.jpeg", "data/張曼玉.jpeg")
    opener.retrieve("https://cfcdn.she.com/media/she.com/2023/03/092803-092803-Screenshot-2023-03-10-at-11.28.01-687x1024.png", "data/關之琳.png")
    opener.retrieve("https://cfcdn.she.com/media/she.com/2023/03/233103-233103-Screenshot-2023-03-10-at-11.31.08-783x1024.png", "data/王祖賢.png")
    opener.retrieve("https://cfcdn.she.com/media/she.com/2023/03/433403-433403-w644-4.jpeg", "data/邱淑貞.jpeg")
    opener.retrieve("https://cfcdn.she.com/media/she.com/2023/03/333503-333503-20210621104344-80d11f2a.jpeg", "data/李嘉欣.jpeg")


import face_recognition

# Often instead of just checking if two faces match or not (True or False), it's helpful to see how similar they are.
# You can do that by using the face_distance function.

# The model was trained in a way that faces with a distance of 0.6 or less should be a match. But if you want to
# be more strict, you can look for a smaller face distance. For example, using a 0.55 cutoff would reduce false
# positive matches at the risk of more false negatives.

# Note: This isn't exactly the same as a "percent match". The scale isn't linear. But you can assume that images with a
# smaller distance are more similar to each other than ones with a larger distance.

# Load some images to compare against
known_encodings = []
known_persons = []
valid_images = [".jpg",".jpeg",".png"]
for f in os.listdir("data"):
    ext = os.path.splitext(f)[1]
    if ext.lower() not in valid_images:
        continue

    # Get the face encodings for the known images
    known_image = face_recognition.load_image_file(os.path.join("data",f))
    face_encoding = face_recognition.face_encodings(known_image)[0]
    known_encodings.append(face_encoding)
    # known_persons.append(os.path.splitext(os.path.basename(f))[0])
    known_persons.append(os.path.basename(f))

import tempfile
import faceSym
from PIL import Image
import numpy as np

def left_right_sim(img):
    tmpf = tempfile.NamedTemporaryFile(delete=False)
    im = Image.fromarray(img)
    im.save(tmpf.name, format='png')
    f = faceSym.FaceSym(tmpf.name)
    _, left, _, _, right, _ = f.get_symmetrized_images(idx=0)
    tmpf.close()
    os.unlink(tmpf.name)
    left_encoding = face_recognition.face_encodings(np.asarray(left))[0]
    right_encoding = face_recognition.face_encodings(np.asarray(right))[0]
    diff = face_recognition.face_distance([left_encoding], right_encoding)
    return 100 * (1 - diff)

import gradio as gr

def greet(image_to_test):
    # # Load a test image and get encondings for it
    # image_to_test = face_recognition.load_image_file(filepath)
    image_to_test_encoding = face_recognition.face_encodings(image_to_test)[0]
    
    # See how far apart the test image is from the known faces
    face_distances = face_recognition.face_distance(known_encodings, image_to_test_encoding)
    idx = face_distances.argmin()
    filepath = known_persons[idx]
    face_distance = face_distances[idx]
    ret = "The most similar person is of {} with score {:.3}".format(
        os.path.splitext(filepath)[0],
        100 * (1 - face_distance))
    img = face_recognition.load_image_file(os.path.join("data", filepath))
    ret += "\n\n \
    The similarity (symmetry score) of \
    left and right face = {:.3}%".format(
        left_right_sim(image_to_test).item())
    return img, ret

iface = gr.Interface(fn=greet, inputs=gr.Image(source="webcam", streaming=True, type="numpy"), outputs=["image", "text"])
iface.launch()