app.py

import cv2
import tempfile
import inspect
from typing import List, Optional, Union
import os
import numpy as np
import torch
import banana_dev as banana
import PIL
from diffusers import AutoencoderKL, DDIMScheduler, DiffusionPipeline, PNDMScheduler, UNet2DConditionModel
from diffusers.pipelines.stable_diffusion import StableDiffusionSafetyChecker
from tqdm.auto import tqdm
from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
import gradio as gr
import random
import base64
from io import BytesIO
import os
import gdown

def download_gdrive_url():
  url = 'https://drive.google.com/u/0/uc?id=1PPO2MCttsmSqyB-vKh5C7SumwFKuhgyj&export=download'
  output = 'haarcascade_frontalface_default.xml'
  gdown.download(url, output, quiet=False)
  
def inpaint(p, init_image, mask_image=None, strength=0.75, guidance_scale=7.5, generator=None, num_samples=1, n_iter=1):
    buffered_init_img = BytesIO()
    buffered_inverted_img = BytesIO()
    init_image.save(buffered_init_img,format="JPEG")
    mask_image.save(buffered_inverted_img,format="JPEG")
    encoded_init_image = base64.b64encode(buffered_init_img.getvalue()).decode('utf-8')
    encoded_inverted_image = base64.b64encode(buffered_inverted_img.getvalue()).decode('utf-8')
    model_inputs = {
      "prompt": "4K UHD professional profile picture of a person wearing a suit for work and posing for a picture, fine details, realistic shaded.",
      "init_image": encoded_init_image,
      "mask_image": encoded_inverted_image,
      "strength": 0.65,
      "guidance_scale": 10,
      "num_inference_steps": 100
    }
    out = banana.run(os.environ.get("API_KEY"), os.environ.get("MODEL_KEY"), model_inputs)
    image_byte_string = out["modelOutputs"][0]["output_image_base64"]
    image_encoded = image_byte_string.encode('utf-8')
    image_bytes = BytesIO(base64.b64decode(image_encoded))
    return_image = Image.open(image_bytes)
    return return_image
    
def identify_face(user_image):
  img = cv2.imread(user_image.name) # read the resized image in cv2
  print(img.shape)
  gray_img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) # convert to grayscale
  download_gdrive_url() #download the haarcascade face recognition stuff
  haar_cascade = cv2.CascadeClassifier('haarcascade_frontalface_default.xml')
  faces_rect = haar_cascade.detectMultiScale(gray_img, scaleFactor=1.1, minNeighbors=9)
  for (x, y, w, h) in faces_rect[:1]:
    mask = np.zeros(img.shape[:2], dtype="uint8")
    print(mask.shape)
    cv2.rectangle(mask, (x, y), (x+w, y+h),  255, -1)
    print(mask.shape)
    inverted_image = cv2.bitwise_not(mask)
    return inverted_image

def sample_images(init_image, mask_image):
  p = "4K UHD professional profile picture of a person wearing a suit for work"
  strength=0.65
  guidance_scale=10
  num_samples = 1
  n_iter = 1

  generator = torch.Generator(device="cuda").manual_seed(random.randint(0, 1000000)) # change the seed to get different results
  all_images = inpaint(p, init_image, mask_image, strength=strength, guidance_scale=guidance_scale, generator=generator, num_samples=num_samples, n_iter=n_iter)
  return all_images

def preprocess_image(image):
    w, h = image.size
    w, h = map(lambda x: x - x % 32, (w, h))  # resize to integer multiple of 32
    image = image.resize((w, h), resample=PIL.Image.LANCZOS)
    image = np.array(image).astype(np.float32) / 255.0
    image = image[None].transpose(0, 3, 1, 2)
    image = torch.from_numpy(image)
    return 2.0 * image - 1.0

def preprocess_mask(mask):
    mask=mask.convert("L")
    w, h = mask.size
    w, h = map(lambda x: x - x % 32, (w, h))  # resize to integer multiple of 32
    mask = mask.resize((w//8, h//8), resample=PIL.Image.NEAREST)
    mask = np.array(mask).astype(np.float32) / 255.0
    mask = np.tile(mask,(4,1,1))
    mask = mask[None].transpose(0, 1, 2, 3)#what does this step do?
    mask = 1 - mask #repaint white, keep black
    mask = torch.from_numpy(mask)
    return mask

# accept an image input 
# trigger the set of functions to occur => identify face, generate mask, save the inverted face mask, sample for the inverted images
# output the sampled images
def main(user_image):
  # accept the image as input
  init_image = PIL.Image.open(user_image).convert("RGB")
  # # resize the image to be (512, 512)
  newsize = (512, 512)
  init_image = init_image.resize(newsize)
  init_image.save(user_image.name) # save the resized image
  ## identify the face + save the inverted mask
  inverted_mask = identify_face(user_image)
  fp = tempfile.NamedTemporaryFile(mode='wb', suffix=".png") 
  cv2.imwrite(fp.name, inverted_mask) # save the inverted image mask
  pil_inverted_mask = PIL.Image.open(fp.name).convert("RGB")
  print("type(init_image): ", type(init_image))
  print("type(pil_inverted_mask): ", type(pil_inverted_mask))
  # sample the new images 
  return sample_images(init_image, pil_inverted_mask)

demo = gr.Interface(main, gr.Image(type="file"), "image")
demo.launch(debug=True)