app.py

import json
import zipfile
import numpy as np
import cv2
import os
import gradio as gr
from deepface import DeepFace
from ultralytics import YOLO
import urllib.request

with open('config.json', 'r') as f:
    config = json.load(f)

FACE_DIST_TRESH = config['FACE_DIST_TRESH']
FACE_DET_TRESH = config['FACE_DET_TRESH']
YOLO_WEIGHTS_URL = config['YOLO_WEIGHTS_URL']

yolo_weights_filename = os.path.basename(YOLO_WEIGHTS_URL)

if not os.path.exists(yolo_weights_filename):
   urllib.request.urlretrieve(YOLO_WEIGHTS_URL, yolo_weights_filename)

model = YOLO(yolo_weights_filename)

def find_distance(base_face, check_face):
  result = DeepFace.verify(base_face, check_face, enforce_detection=False)
  return result['distance']

def find_faces(image):
  outputs = model(image)
  faces = []
  for box in outputs[0].boxes:
    if float(box.conf) >= FACE_DET_TRESH:
      x, y, w, h = [int(coord) for coord in box.xywh[0]]
      x_center, y_center = x + w / 2, y + h / 2
      x1 = int(x_center - w)
      y1 = int(y_center - h)
      crop_img = image[y1:y1+h, x1:x1+w]
      faces.append(crop_img)
  return faces
  
def load_images_from_zip(zip_path):
    images = []
    with zipfile.ZipFile(zip_path, 'r') as zip_file:
        for file_name in zip_file.namelist():
            with zip_file.open(file_name) as file:
                img_bytes = file.read()
                img = cv2.imdecode(np.frombuffer(img_bytes, np.uint8), cv2.IMREAD_COLOR)
                if img is not None:
                    images.append(img)
    return images

def create_image(images):
    table_width = 800
    row_height = 100
    margin = 10
    text_margin = 20
    id_col_width = 100

    font = cv2.FONT_HERSHEY_SIMPLEX
    font_scale = 0.5
    color = (255, 255, 255)
    thickness = 2

    table_height = text_margin + margin + (row_height + margin) * len(images)

    table = np.zeros((table_height, table_width, 3), np.uint8)

    id_x = 10
    img_x = id_col_width + 10
    y = text_margin

    cv2.putText(table, 'Image ID', (id_x, y), font, font_scale, color, thickness)
    cv2.putText(table, 'Face', (img_x, y), font, font_scale, color, thickness)

    y += margin

    for i, img in enumerate(images):
        height, width = img.shape[:2]
        new_width = int(width * row_height / height)
        if img_x + new_width > table_width:
            new_width = table_width - img_x
        img_resized = cv2.resize(img, (new_width, row_height))

        cv2.putText(table, str(i), (id_x, y + margin), font, font_scale, color, thickness)
        table[y:y+row_height, img_x:img_x+new_width] = img_resized

        y += row_height + margin

    for col in range(table.shape[1]-1, -1, -1):
        if not np.any(table[:, col]):
            continue
        else:
            break
    table_cropped = table[:, :col+1+id_x]

    return table_cropped

def check(avatars_zip, photos_zip):
    avatars = [cv2.cvtColor(img, cv2.COLOR_RGB2BGR) for img in load_images_from_zip(avatars_zip.name)]
    photos = [cv2.cvtColor(img, cv2.COLOR_RGB2BGR) for img in load_images_from_zip(photos_zip.name)]

    input_avatars_faces = [find_faces(avatar) for avatar in avatars]
    input_avatars_faces = [face for faces in input_avatars_faces for face in faces]

    avatars_faces_count = len(input_avatars_faces)

    not_found_faces = []

    for photo in photos:
        input_faces = find_faces(photo)
        for input_face in input_faces:
            for i in range(avatars_faces_count):
                distance = find_distance(input_avatars_faces[i], input_face)
                if distance <= FACE_DIST_TRESH:
                    break
                elif i + 1 == avatars_faces_count:
                    not_found_faces.append(input_face)

    return create_image(not_found_faces)

title = '<h1 style="text-align:center">SquadDetective</h1>'
logo = '<center><img src="https://i.ibb.co/C0BH40g/logo.png" width="300" height="300" alt="SquadDetective logo"></center>'

with gr.Blocks(theme='soft', title='SquadDetective') as blocks:
    gr.HTML(title)
    gr.HTML(logo)
    gr.Markdown('SquadDetective is a service that helps sports teams to identify unclaimed players by comparing their faces to photos taken during matches. By using state-of-the-art facial recognition technology, SquadDetective can quickly and accurately match the faces of players in photos to a database of registered players, allowing teams to quickly identify any unclaimed players and take appropriate action. With SquadDetective, sports teams can ensure that all players are properly registered and eligible to play, helping to avoid potential penalties and other issues.')
    with gr.Row():
        avatars = gr.inputs.File(label='Avatar photos (zip)')
        photos = gr.inputs.File(label='Photos to be processed (zip)')
    inputs = [avatars, photos]
    process_button = gr.Button('Process')
    outputs=gr.outputs.Image(type='numpy', label='Report')
    process_button.click(fn=check, inputs=inputs, outputs=outputs)

blocks.launch()