Solution for submission 175118

!pip install aicrowd-cli

# Make sure to re-run below code whenever you restart colab notebook
%load_ext aicrowd.magic

# Logging in from our AIcrowd account. Make sure you have accepted the puzzle rules before logging in!  

%aicrowd login

# Creating a new data directory and downloading the dataset 

!rm -rf data
!mkdir data
%aicrowd ds dl -c face-recognition -o data

# unzipping the data

!unzip data/data.zip -d /content/data > /dev/null

import pandas as pd
import os
import numpy as np
import random
from tqdm.notebook import tqdm
from google.colab.patches import cv2_imshow
import cv2

# Getting all image ids from a folder

image_ids = os.listdir("data/missing")
len(image_ids)

# We can also split all the faces in the target image to convert them into individual faces images

sample_target_faces = []


def get_target_face(face_no, target_image):


  # Top-Left x, y corrdinates of the specific face 
  x, y = (int(face_no[0]))*216, (int(face_no[1]))*216

  target_face = target_image[x:x+216, y:y+216]

  return target_face

!pip install face-recognition

import cv2
import numpy as np
import math
import face_recognition

predictions = {"ImageID":[], "target":[]}

# Find for every missing image a target image
for img_id in tqdm(image_ids):

  # Face no with minimum error
  min_error_face_no = 0
  min_distance = 100000000

  try:
    missing_image = face_recognition.load_image_file("data/missing/"+ img_id)
    missing_encoding_small = face_recognition.face_encodings(missing_image, num_jitters=2, model='small')[0]
    missing_encoding_large = face_recognition.face_encodings(missing_image, num_jitters=2, model='large')[0]

    target_image_original = cv2.imread("data/target/"+ img_id, 1)
    target_distances = []
    for face_no in range(100):
      # Getting the specific face from the target image
      face_no = str(face_no)
      face_no = face_no.zfill(2)

      single_target_face = get_target_face(face_no, target_image_original) # get one single image
      cv2.imwrite("tmp_target_face.jpg", single_target_face) # save cv image

      try:
        target_image = face_recognition.load_image_file("tmp_target_face.jpg") # load image
        target_encodings_small = face_recognition.face_encodings(target_image, num_jitters=2, model='small')
        target_encodings_large = face_recognition.face_encodings(target_image, num_jitters=2, model='large')

        # It could be that there are multiple faces in one image
        # We are using both: The small and the large model to find the best target over all
        min_distance_small = min_distance
        for target_encoding in target_encodings_small:
          face_distance = face_recognition.face_distance([missing_encoding_small], target_encoding)
          # We want the smallest distance in this picture for small model
          if face_distance < min_distance_small:
            min_distance_small = face_distance

        min_distance_large = min_distance
        for target_encoding in target_encodings_large:
          face_distance = face_recognition.face_distance([missing_encoding_large], target_encoding)
          # We want the smallest distance in this picture for large model
          if face_distance < min_distance_large:
            min_distance_large = face_distance

        # Add distance mean to target_distance list
        target_distances.append( (min_distance_small + min_distance_large) /2 )


      except:
        print("The facereconition went wrong for a target image")
        target_image_tmp = cv2.imread("tmp_target_face.jpg")
        cv2_imshow(target_image_tmp)

    # The target face with the minimal distance is the searched for
    min_error_face_no = str(target_distances.index(min(target_distances)))
    min_error_face_no = min_error_face_no.zfill(2)
  except:
    print("The facereconition went wrong for a missing image")
    
  predictions['ImageID'].append(img_id.replace(".jpg", ""))
  predictions['target'].append(min_error_face_no)

# ---- Visualize above
for i in range(len(predictions['ImageID'][:2])):
  print("Missing: "+ predictions['ImageID'][i])
  file_name = predictions['ImageID'][i] + ".jpg"
  missing_image = cv2.imread(os.path.join("data/missing", file_name), 0)
  cv2_imshow(missing_image)

  print()
  print("--------Target--------")
  print(predictions['target'][i])
  target_image = cv2.imread(os.path.join("data/target", file_name), 0)
  sample_target_face = get_target_face(str(predictions['target'][i]), target_image)
  cv2_imshow(sample_target_face)

  print("----------------------")
  print()

submission = pd.DataFrame(predictions)
submission

# Saving the predictions
!rm -rf assets
!mkdir assets
submission.to_csv(os.path.join("assets", "submission.csv"), index=False)

%aicrowd notebook submit -c face-recognition -a assets --no-verify