|
import os |
|
import sys |
|
from torchvision.transforms import functional |
|
sys.modules["torchvision.transforms.functional_tensor"] = functional |
|
|
|
from basicsr.archs.srvgg_arch import SRVGGNetCompact |
|
from gfpgan.utils import GFPGANer |
|
from realesrgan.utils import RealESRGANer |
|
|
|
import torch |
|
import cv2 |
|
import gradio as gr |
|
|
|
|
|
def download_upscaler_models(): |
|
if not os.path.exists('realesr-general-x4v3.pth'): |
|
os.system("wget https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-x4v3.pth -P .") |
|
if not os.path.exists('GFPGANv1.4.pth'): |
|
os.system("wget https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.4.pth -P .") |
|
|
|
|
|
def init_upscaler_model(): |
|
model = SRVGGNetCompact(num_in_ch=3, num_out_ch=3, num_feat=64, num_conv=32, upscale=4, act_type='prelu') |
|
model_path = 'realesr-general-x4v3.pth' |
|
half = True if torch.cuda.is_available() else False |
|
upsampler = RealESRGANer(scale=4, model_path=model_path, model=model, tile=0, tile_pad=10, pre_pad=0, half=half) |
|
|
|
return upsampler |
|
|
|
|
|
def upscale_image(img): |
|
try: |
|
upsampler = init_upscaler_model() |
|
|
|
|
|
img = cv2.imread(img, cv2.IMREAD_UNCHANGED) |
|
if len(img.shape) == 3 and img.shape[2] == 4: |
|
img_mode = 'RGBA' |
|
elif len(img.shape) == 2: |
|
img_mode = None |
|
img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR) |
|
else: |
|
img_mode = None |
|
|
|
|
|
h, w = img.shape[0:2] |
|
if h < 300: |
|
img = cv2.resize(img, (w * 2, h * 2), interpolation=cv2.INTER_LANCZOS4) |
|
|
|
|
|
face_enhancer = GFPGANer( |
|
model_path='GFPGANv1.4.pth', |
|
upscale=2, |
|
arch='clean', |
|
channel_multiplier=2, |
|
bg_upsampler=upsampler |
|
) |
|
|
|
|
|
try: |
|
_, _, output = face_enhancer.enhance(img, has_aligned=False, only_center_face=False, paste_back=True) |
|
except RuntimeError as error: |
|
print('์ค๋ฅ', error) |
|
return None |
|
|
|
|
|
output = cv2.cvtColor(output, cv2.COLOR_BGR2RGB) |
|
return output |
|
except Exception as error: |
|
print('๊ธ๋ก๋ฒ ์์ธ', error) |
|
return None |
|
|
|
|
|
def create_upscaler_tab(): |
|
|
|
download_upscaler_models() |
|
|
|
with gr.Tab("์ด๋ฏธ์ง ์
์ค์ผ์ผ๋ฌ"): |
|
gr.Markdown( |
|
""" |
|
# ์ด๋ฏธ์ง ์
์ค์ผ์ผ๋ฌ |
|
|
|
์
๋ก๋ํ ์ด๋ฏธ์ง๋ฅผ ๊ณ ํด์๋๋ก ๋ณํํฉ๋๋ค. ํนํ ์ธ๋ฌผ ์ฌ์ง์ ์ผ๊ตด ๋ํ
์ผ์ ํฅ์์ํต๋๋ค. |
|
""" |
|
) |
|
|
|
with gr.Row(): |
|
with gr.Column(): |
|
input_image = gr.Image(type="filepath", label="์
๋ ฅ ์ด๋ฏธ์ง") |
|
upscale_btn = gr.Button("์ด๋ฏธ์ง ์
์ค์ผ์ผ", variant="primary") |
|
|
|
with gr.Column(): |
|
output_image = gr.Image(type="numpy", label="๊ฒฐ๊ณผ ์ด๋ฏธ์ง") |
|
|
|
upscale_btn.click( |
|
fn=upscale_image, |
|
inputs=[input_image], |
|
outputs=[output_image] |
|
) |
|
|
|
import os |
|
import tempfile |
|
from PIL import Image |
|
import gradio as gr |
|
import logging |
|
import re |
|
from io import BytesIO |
|
import time |
|
|
|
|
|
import sys |
|
from torchvision.transforms import functional |
|
sys.modules["torchvision.transforms.functional_tensor"] = functional |
|
|
|
from basicsr.archs.srvgg_arch import SRVGGNetCompact |
|
from gfpgan.utils import GFPGANer |
|
from realesrgan.utils import RealESRGANer |
|
|
|
import torch |
|
import cv2 |
|
|
|
from google import genai |
|
from google.genai import types |
|
|
|
|
|
from dotenv import load_dotenv |
|
load_dotenv() |
|
|
|
|
|
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s') |
|
logger = logging.getLogger(__name__) |
|
|
|
def save_binary_file(file_name, data): |
|
with open(file_name, "wb") as f: |
|
f.write(data) |
|
|
|
def translate_prompt_to_english(prompt): |
|
""" |
|
์
๋ ฅ๋ ํ๋กฌํํธ์ ํ๊ธ์ด ํฌํจ๋์ด ์์ผ๋ฉด Geminiโ2.0โflash ๋ชจ๋ธ์ ์ฌ์ฉํ์ฌ ์์ด๋ก ๋ฒ์ญํฉ๋๋ค. |
|
ํ๊ธ์ด ์์ผ๋ฉด ์๋ณธ ํ๋กฌํํธ๋ฅผ ๊ทธ๋๋ก ๋ฐํํฉ๋๋ค. |
|
์ค์: #1, #2, #3 ํ๊ทธ๋ ๋ฒ์ญ ์ ํ์ ๋ฐ๋์ ๋ณด์กด๋์ด์ผ ํฉ๋๋ค. |
|
""" |
|
if not re.search("[๊ฐ-ํฃ]", prompt): |
|
return prompt |
|
|
|
|
|
prompt = prompt.replace("#1", "IMAGE_TAG_ONE") |
|
prompt = prompt.replace("#2", "IMAGE_TAG_TWO") |
|
prompt = prompt.replace("#3", "IMAGE_TAG_THREE") |
|
|
|
try: |
|
api_key = os.environ.get("GEMINI_API_KEY") |
|
if not api_key: |
|
logger.error("Gemini API ํค๊ฐ ์ค์ ๋์ง ์์์ต๋๋ค.") |
|
|
|
prompt = prompt.replace("IMAGE_TAG_ONE", "#1") |
|
prompt = prompt.replace("IMAGE_TAG_TWO", "#2") |
|
prompt = prompt.replace("IMAGE_TAG_THREE", "#3") |
|
return prompt |
|
|
|
client = genai.Client(api_key=api_key) |
|
translation_prompt = f""" |
|
Translate the following Korean text to English: |
|
|
|
{prompt} |
|
|
|
IMPORTANT: The tokens IMAGE_TAG_ONE, IMAGE_TAG_TWO, and IMAGE_TAG_THREE are special tags |
|
and must be preserved exactly as is in your translation. Do not translate these tokens. |
|
""" |
|
|
|
logger.info(f"Translation prompt: {translation_prompt}") |
|
response = client.models.generate_content( |
|
model="gemini-2.0-flash", |
|
contents=[translation_prompt], |
|
config=types.GenerateContentConfig( |
|
response_modalities=['Text'], |
|
temperature=0.2, |
|
top_p=0.95, |
|
top_k=40, |
|
max_output_tokens=512 |
|
) |
|
) |
|
|
|
translated_text = "" |
|
for part in response.candidates[0].content.parts: |
|
if hasattr(part, 'text') and part.text: |
|
translated_text += part.text |
|
|
|
if translated_text.strip(): |
|
|
|
translated_text = translated_text.replace("IMAGE_TAG_ONE", "#1") |
|
translated_text = translated_text.replace("IMAGE_TAG_TWO", "#2") |
|
translated_text = translated_text.replace("IMAGE_TAG_THREE", "#3") |
|
logger.info(f"Translated text: {translated_text.strip()}") |
|
return translated_text.strip() |
|
else: |
|
logger.warning("๋ฒ์ญ ๊ฒฐ๊ณผ๊ฐ ์์ต๋๋ค. ์๋ณธ ํ๋กฌํํธ ์ฌ์ฉ") |
|
|
|
prompt = prompt.replace("IMAGE_TAG_ONE", "#1") |
|
prompt = prompt.replace("IMAGE_TAG_TWO", "#2") |
|
prompt = prompt.replace("IMAGE_TAG_THREE", "#3") |
|
return prompt |
|
except Exception as e: |
|
logger.exception("๋ฒ์ญ ์ค ์ค๋ฅ ๋ฐ์:") |
|
|
|
prompt = prompt.replace("IMAGE_TAG_ONE", "#1") |
|
prompt = prompt.replace("IMAGE_TAG_TWO", "#2") |
|
prompt = prompt.replace("IMAGE_TAG_THREE", "#3") |
|
return prompt |
|
|
|
def preprocess_prompt(prompt, image1, image2, image3): |
|
""" |
|
ํ๋กฌํํธ๋ฅผ ์ฒ๋ฆฌํ๊ณ ๊ธฐ๋ฅ ๋ช
๋ น์ ํด์ |
|
""" |
|
has_img1 = image1 is not None |
|
has_img2 = image2 is not None |
|
has_img3 = image3 is not None |
|
|
|
if "#1" in prompt and not has_img1: |
|
prompt = prompt.replace("#1", "์ฒซ ๋ฒ์งธ ์ด๋ฏธ์ง(์์)") |
|
else: |
|
prompt = prompt.replace("#1", "์ฒซ ๋ฒ์งธ ์ด๋ฏธ์ง") |
|
|
|
if "#2" in prompt and not has_img2: |
|
prompt = prompt.replace("#2", "๋ ๋ฒ์งธ ์ด๋ฏธ์ง(์์)") |
|
else: |
|
prompt = prompt.replace("#2", "๋ ๋ฒ์งธ ์ด๋ฏธ์ง") |
|
|
|
if "#3" in prompt and not has_img3: |
|
prompt = prompt.replace("#3", "์ธ ๋ฒ์งธ ์ด๋ฏธ์ง(์์)") |
|
else: |
|
prompt = prompt.replace("#3", "์ธ ๋ฒ์งธ ์ด๋ฏธ์ง") |
|
|
|
if "1. ์ด๋ฏธ์ง ๋ณ๊ฒฝ" in prompt: |
|
desc_match = re.search(r'#1์ "(.*?)"์ผ๋ก ๋ฐ๊ฟ๋ผ', prompt) |
|
if desc_match: |
|
description = desc_match.group(1) |
|
prompt = f"์ฒซ ๋ฒ์งธ ์ด๋ฏธ์ง๋ฅผ {description}์ผ๋ก ๋ณ๊ฒฝํด์ฃผ์ธ์. ์๋ณธ ์ด๋ฏธ์ง์ ์ฃผ์ ๋ด์ฉ์ ์ ์งํ๋ ์๋ก์ด ์คํ์ผ๊ณผ ๋ถ์๊ธฐ๋ก ์ฌํด์ํด์ฃผ์ธ์." |
|
else: |
|
prompt = "์ฒซ ๋ฒ์งธ ์ด๋ฏธ์ง๋ฅผ ์ฐฝ์์ ์ผ๋ก ๋ณํํด์ฃผ์ธ์. ๋ ์์ํ๊ณ ์์ ์ ์ธ ๋ฒ์ ์ผ๋ก ๋ง๋ค์ด์ฃผ์ธ์." |
|
|
|
elif "2. ๊ธ์์ง์ฐ๊ธฐ" in prompt: |
|
text_match = re.search(r'#1์์ "(.*?)"๋ฅผ ์ง์๋ผ', prompt) |
|
if text_match: |
|
text_to_remove = text_match.group(1) |
|
prompt = f"์ฒซ ๋ฒ์งธ ์ด๋ฏธ์ง์์ '{text_to_remove}' ํ
์คํธ๋ฅผ ์ฐพ์ ์์ฐ์ค๋ฝ๊ฒ ์ ๊ฑฐํด์ฃผ์ธ์. ํ
์คํธ๊ฐ ์๋ ๋ถ๋ถ์ ๋ฐฐ๊ฒฝ๊ณผ ์กฐํ๋กญ๊ฒ ์ฑ์์ฃผ์ธ์." |
|
else: |
|
prompt = "์ฒซ ๋ฒ์งธ ์ด๋ฏธ์ง์์ ๋ชจ๋ ํ
์คํธ๋ฅผ ์ฐพ์ ์์ฐ์ค๋ฝ๊ฒ ์ ๊ฑฐํด์ฃผ์ธ์. ๊น๋ํ ์ด๋ฏธ์ง๋ก ๋ง๋ค์ด์ฃผ์ธ์." |
|
|
|
elif "4. ์ท๋ฐ๊พธ๊ธฐ" in prompt: |
|
prompt = "์ฒซ ๋ฒ์งธ ์ด๋ฏธ์ง์ ์ธ๋ฌผ ์์์ ๋ ๋ฒ์งธ ์ด๋ฏธ์ง์ ์์์ผ๋ก ๋ณ๊ฒฝํด์ฃผ์ธ์. ์์์ ์คํ์ผ๊ณผ ์์์ ๋ ๋ฒ์งธ ์ด๋ฏธ์ง๋ฅผ ๋ฐ๋ฅด๋, ์ ์ฒด ๋น์จ๊ณผ ํฌ์ฆ๋ ์ฒซ ๋ฒ์งธ ์ด๋ฏธ์ง๋ฅผ ์ ์งํด์ฃผ์ธ์." |
|
|
|
elif "5. ๋ฐฐ๊ฒฝ๋ฐ๊พธ๊ธฐ" in prompt: |
|
prompt = "์ฒซ ๋ฒ์งธ ์ด๋ฏธ์ง์ ๋ฐฐ๊ฒฝ์ ๋ ๋ฒ์งธ ์ด๋ฏธ์ง์ ๋ฐฐ๊ฒฝ์ผ๋ก ๋ณ๊ฒฝํด์ฃผ์ธ์. ์ฒซ ๋ฒ์งธ ์ด๋ฏธ์ง์ ์ฃผ์ ํผ์ฌ์ฒด๋ ์ ์งํ๊ณ , ๋ ๋ฒ์งธ ์ด๋ฏธ์ง์ ๋ฐฐ๊ฒฝ๊ณผ ์กฐํ๋กญ๊ฒ ํฉ์ฑํด์ฃผ์ธ์." |
|
|
|
elif "6. ์ด๋ฏธ์ง ํฉ์ฑ(์ํํฌํจ)" in prompt: |
|
prompt = "์ฒซ ๋ฒ์งธ ์ด๋ฏธ์ง์ ๋ ๋ฒ์งธ ์ด๋ฏธ์ง(๋๋ ์ธ ๋ฒ์งธ ์ด๋ฏธ์ง)๋ฅผ ์์ฐ์ค๋ฝ๊ฒ ํฉ์ฑํด์ฃผ์ธ์. ๋ชจ๋ ์ด๋ฏธ์ง์ ์ฃผ์ ์์๋ฅผ ํฌํจํ๊ณ , ํนํ ์ํ์ด ๋๋ณด์ด๋๋ก ์กฐํ๋กญ๊ฒ ํตํฉํด์ฃผ์ธ์." |
|
|
|
prompt += " ์ด๋ฏธ์ง๋ฅผ ์์ฑํด์ฃผ์ธ์. ์ด๋ฏธ์ง์ ํ
์คํธ๋ ๊ธ์๋ฅผ ํฌํจํ์ง ๋ง์ธ์." |
|
return prompt |
|
|
|
def generate_with_images(prompt, images, variation_index=0): |
|
""" |
|
API ํธ์ถ์ ํตํด ์ด๋ฏธ์ง๋ฅผ ์์ฑํ๊ณ ๊ฒฐ๊ณผ ์ด๋ฏธ์ง๋ฅผ ๋ฐํํฉ๋๋ค. |
|
variation_index๋ก ๋ค์ํ ๋ณํ๋ฅผ ์ค๋๋ค. |
|
""" |
|
try: |
|
api_key = os.environ.get("GEMINI_API_KEY") |
|
if not api_key: |
|
return None, "API ํค๊ฐ ์ค์ ๋์ง ์์์ต๋๋ค. ํ๊ฒฝ๋ณ์๋ฅผ ํ์ธํด์ฃผ์ธ์." |
|
|
|
client = genai.Client(api_key=api_key) |
|
logger.info(f"Gemini API ์์ฒญ ์์ - ํ๋กฌํํธ: {prompt}, ๋ณํ ์ธ๋ฑ์ค: {variation_index}") |
|
|
|
|
|
variation_suffixes = [ |
|
" Create this as the first variation. Do not add any text, watermarks, or labels to the image.", |
|
" Create this as the second variation with more vivid colors. Do not add any text, watermarks, or labels to the image.", |
|
" Create this as the third variation with a more creative style. Do not add any text, watermarks, or labels to the image.", |
|
" Create this as the fourth variation with enhanced details. Do not add any text, watermarks, or labels to the image." |
|
] |
|
|
|
if variation_index < len(variation_suffixes): |
|
prompt = prompt + variation_suffixes[variation_index] |
|
else: |
|
prompt = prompt + " Do not add any text, watermarks, or labels to the image." |
|
|
|
contents = [prompt] |
|
for idx, img in enumerate(images, 1): |
|
if img is not None: |
|
contents.append(img) |
|
logger.info(f"์ด๋ฏธ์ง #{idx} ์ถ๊ฐ๋จ") |
|
|
|
response = client.models.generate_content( |
|
model="gemini-2.0-flash-exp-image-generation", |
|
contents=contents, |
|
config=types.GenerateContentConfig( |
|
response_modalities=['Text', 'Image'], |
|
temperature=1, |
|
top_p=0.95, |
|
top_k=40, |
|
max_output_tokens=8192 |
|
) |
|
) |
|
|
|
with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as tmp: |
|
temp_path = tmp.name |
|
result_text = "" |
|
image_found = False |
|
for part in response.candidates[0].content.parts: |
|
if hasattr(part, 'text') and part.text: |
|
result_text += part.text |
|
logger.info(f"์๋ต ํ
์คํธ: {part.text}") |
|
elif hasattr(part, 'inline_data') and part.inline_data: |
|
save_binary_file(temp_path, part.inline_data.data) |
|
image_found = True |
|
logger.info("์๋ต์์ ์ด๋ฏธ์ง ์ถ์ถ ์ฑ๊ณต") |
|
if not image_found: |
|
return None, f"API์์ ์ด๋ฏธ์ง๋ฅผ ์์ฑํ์ง ๋ชปํ์ต๋๋ค. ์๋ต ํ
์คํธ: {result_text}" |
|
result_img = Image.open(temp_path) |
|
if result_img.mode == "RGBA": |
|
result_img = result_img.convert("RGB") |
|
return result_img, f"์ด๋ฏธ์ง๊ฐ ์ฑ๊ณต์ ์ผ๋ก ์์ฑ๋์์ต๋๋ค. {result_text}" |
|
except Exception as e: |
|
logger.exception("์ด๋ฏธ์ง ์์ฑ ์ค ์ค๋ฅ ๋ฐ์:") |
|
return None, f"์ค๋ฅ ๋ฐ์: {str(e)}" |
|
|
|
def process_images_with_prompt(image1, image2, image3, prompt, variation_index=0, max_retries=3): |
|
""" |
|
3๊ฐ์ ์ด๋ฏธ์ง์ ํ๋กฌํํธ๋ฅผ ์ฒ๋ฆฌํ์ฌ ์ต์ข
์์ด ํ๋กฌํํธ(final_prompt)๋ฅผ ์์ฑํ ํ, |
|
API๋ฅผ ํธ์ถํ์ฌ ๊ฒฐ๊ณผ ์ด๋ฏธ์ง๋ฅผ ๋ฐํํฉ๋๋ค. ์๋ฌ ๋ฐ์ ์ ์ต๋ max_retries ํ์๋งํผ ์ฌ์๋ํฉ๋๋ค. |
|
""" |
|
retry_count = 0 |
|
last_error = None |
|
|
|
while retry_count < max_retries: |
|
try: |
|
images = [image1, image2, image3] |
|
valid_images = [img for img in images if img is not None] |
|
if not valid_images: |
|
return None, "์ ์ด๋ ํ๋์ ์ด๋ฏธ์ง๋ฅผ ์
๋ก๋ํด์ฃผ์ธ์.", "" |
|
|
|
if prompt and prompt.strip(): |
|
processed_prompt = preprocess_prompt(prompt, image1, image2, image3) |
|
if re.search("[๊ฐ-ํฃ]", processed_prompt): |
|
final_prompt = translate_prompt_to_english(processed_prompt) |
|
else: |
|
final_prompt = processed_prompt |
|
else: |
|
if len(valid_images) == 1: |
|
final_prompt = "Please creatively transform this image into a more vivid and artistic version. Do not include any text or watermarks in the generated image." |
|
logger.info("Default prompt generated for single image") |
|
elif len(valid_images) == 2: |
|
final_prompt = "Please seamlessly composite these two images, integrating their key elements harmoniously into a single image. Do not include any text or watermarks in the generated image." |
|
logger.info("Default prompt generated for two images") |
|
else: |
|
final_prompt = "Please creatively composite these three images, combining their main elements into a cohesive and natural scene. Do not include any text or watermarks in the generated image." |
|
logger.info("Default prompt generated for three images") |
|
|
|
result_img, status = generate_with_images(final_prompt, valid_images, variation_index) |
|
if result_img is not None: |
|
return result_img, status, final_prompt |
|
else: |
|
last_error = status |
|
retry_count += 1 |
|
logger.warning(f"์ด๋ฏธ์ง ์์ฑ ์คํจ, ์ฌ์๋ {retry_count}/{max_retries}: {status}") |
|
time.sleep(1) |
|
except Exception as e: |
|
last_error = str(e) |
|
retry_count += 1 |
|
logger.exception(f"์ด๋ฏธ์ง ์ฒ๋ฆฌ ์ค ์ค๋ฅ ๋ฐ์, ์ฌ์๋ {retry_count}/{max_retries}:") |
|
time.sleep(1) |
|
|
|
return None, f"์ต๋ ์ฌ์๋ ํ์({max_retries}ํ) ์ด๊ณผ ํ ์คํจ: {last_error}", prompt |
|
|
|
def generate_multiple_images(image1, image2, image3, prompt, progress=gr.Progress()): |
|
""" |
|
์ฌ๋ฌ ๊ฐ์ ์ด๋ฏธ์ง๋ฅผ ์ฐจ๋ก๋๋ก ์์ฑํฉ๋๋ค. |
|
""" |
|
results = [] |
|
statuses = [] |
|
prompts = [] |
|
|
|
num_images = 4 |
|
max_retries = 3 |
|
|
|
progress(0, desc="์ด๋ฏธ์ง ์์ฑ ์ค๋น ์ค...") |
|
|
|
for i in range(num_images): |
|
progress((i / num_images), desc=f"{i+1}/{num_images} ์ด๋ฏธ์ง ์์ฑ ์ค...") |
|
result_img, status, final_prompt = process_images_with_prompt(image1, image2, image3, prompt, i, max_retries) |
|
|
|
if result_img is not None: |
|
results.append(result_img) |
|
statuses.append(f"์ด๋ฏธ์ง #{i+1}: {status}") |
|
prompts.append(f"์ด๋ฏธ์ง #{i+1}: {final_prompt}") |
|
else: |
|
results.append(None) |
|
statuses.append(f"์ด๋ฏธ์ง #{i+1} ์์ฑ ์คํจ: {status}") |
|
prompts.append(f"์ด๋ฏธ์ง #{i+1}: {final_prompt}") |
|
|
|
time.sleep(1) |
|
|
|
progress(1.0, desc="์ด๋ฏธ์ง ์์ฑ ์๋ฃ!") |
|
|
|
while len(results) < 4: |
|
results.append(None) |
|
|
|
combined_status = "\n".join(statuses) |
|
combined_prompts = "\n".join(prompts) |
|
|
|
return results[0], results[1], results[2], results[3], combined_status, combined_prompts |
|
|
|
|
|
def download_upscaler_models(): |
|
if not os.path.exists('realesr-general-x4v3.pth'): |
|
os.system("wget https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-x4v3.pth -P .") |
|
if not os.path.exists('GFPGANv1.4.pth'): |
|
os.system("wget https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.4.pth -P .") |
|
|
|
def upscale_image(img): |
|
try: |
|
|
|
model = SRVGGNetCompact(num_in_ch=3, num_out_ch=3, num_feat=64, num_conv=32, upscale=4, act_type='prelu') |
|
model_path = 'realesr-general-x4v3.pth' |
|
half = True if torch.cuda.is_available() else False |
|
upsampler = RealESRGANer(scale=4, model_path=model_path, model=model, tile=0, tile_pad=10, pre_pad=0, half=half) |
|
|
|
|
|
img = cv2.imread(img, cv2.IMREAD_UNCHANGED) |
|
if len(img.shape) == 3 and img.shape[2] == 4: |
|
img_mode = 'RGBA' |
|
elif len(img.shape) == 2: |
|
img_mode = None |
|
img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR) |
|
else: |
|
img_mode = None |
|
|
|
|
|
h, w = img.shape[0:2] |
|
if h < 300: |
|
img = cv2.resize(img, (w * 2, h * 2), interpolation=cv2.INTER_LANCZOS4) |
|
|
|
|
|
face_enhancer = GFPGANer( |
|
model_path='GFPGANv1.4.pth', |
|
upscale=2, |
|
arch='clean', |
|
channel_multiplier=2, |
|
bg_upsampler=upsampler |
|
) |
|
|
|
|
|
try: |
|
_, _, output = face_enhancer.enhance(img, has_aligned=False, only_center_face=False, paste_back=True) |
|
except RuntimeError as error: |
|
print('์ค๋ฅ', error) |
|
return None |
|
|
|
|
|
output = cv2.cvtColor(output, cv2.COLOR_BGR2RGB) |
|
return output |
|
except Exception as error: |
|
print('๊ธ๋ก๋ฒ ์์ธ', error) |
|
return None |
|
|
|
|
|
with gr.Blocks() as demo: |
|
gr.HTML( |
|
""" |
|
<div style="text-align: center; margin-bottom: 1rem;"> |
|
<h1>์ด์ปค๋จธ์ค์ฉ ์ด๋ฏธ์ง ์์ฑ ๋ฐ ํธ์ง ๋๊ตฌ</h1> |
|
<p>์ด๋ฏธ์ง๋ฅผ ์
๋ก๋ํ๊ณ ์์ ๋ฅผ ์ฐธ๊ณ ํ์ฌ ํ๋กฌํํธ๋ฅผ ์์ ํ๊ฑฐ๋ ์ด๋ฏธ์ง ์
์ค์ผ์ผ ๊ธฐ๋ฅ์ ์ด์ฉํ์ธ์.</p> |
|
</div> |
|
""" |
|
) |
|
|
|
with gr.Tabs(): |
|
|
|
with gr.Tab("์ด๋ฏธ์ง ์์ฑ๊ธฐ"): |
|
with gr.Row(): |
|
|
|
with gr.Column(scale=1): |
|
with gr.Row(): |
|
image1_input = gr.Image(type="pil", label="#1", image_mode="RGB", height=300, width=200) |
|
image2_input = gr.Image(type="pil", label="#2", image_mode="RGB", height=300, width=200) |
|
image3_input = gr.Image(type="pil", label="#3", image_mode="RGB", height=300, width=200) |
|
prompt_input = gr.Textbox( |
|
lines=3, |
|
placeholder="ํ๋กฌํํธ๋ฅผ ์
๋ ฅํ๊ฑฐ๋ ๋น์๋๋ฉด ์๋ ํฉ์ฑ๋ฉ๋๋ค.", |
|
label="ํ๋กฌํํธ (์ ํ ์ฌํญ)" |
|
) |
|
with gr.Row(): |
|
image_change_btn1 = gr.Button("์ด๋ฏธ์ง ๋ณ๊ฒฝ-1") |
|
image_change_btn2 = gr.Button("์ด๋ฏธ์ง ๋ณ๊ฒฝ-2") |
|
text_remove_btn = gr.Button("๊ธ์ ์ง์ฐ๊ธฐ") |
|
text_change_btn = gr.Button("๊ธ์ ๋ณ๊ฒฝํ๊ธฐ") |
|
clothes_change_btn1 = gr.Button("๊ฐ์ ์ํ์ฐฉ์ฉ-1") |
|
clothes_change_btn2 = gr.Button("๊ฐ์ ์ํ์ฐฉ์ฉ-2") |
|
holding_product_btn = gr.Button("์ํ๋ค๊ณ ์๊ธฐ") |
|
background_change_btn = gr.Button("๋ฐฐ๊ฒฝ ๋ฐ๊พธ๊ธฐ") |
|
composite_product_btn = gr.Button("๋ถ๋ถ ์ง์ฐ๊ธฐ") |
|
submit_btn = gr.Button("์ด๋ฏธ์ง ์์ฑ (4์ฅ)", variant="primary") |
|
|
|
gr.Markdown( |
|
""" |
|
### ์ฌ์ฉ ๋ฐฉ๋ฒ: |
|
|
|
1. **์๋ ํฉ์ฑ**: ์ด๋ฏธ์ง๋ฅผ ์
๋ก๋ํ๊ณ ํ๋กฌํํธ๋ฅผ ๋น์๋๋ฉด ์๋์ผ๋ก ํฉ์ฑ๋ฉ๋๋ค. |
|
2. **์ด๋ฏธ์ง ์ฐธ์กฐ**: #1, #2, #3์ผ๋ก ๊ฐ ์ด๋ฏธ์ง๋ฅผ ์ฐธ์กฐํ ์ ์์ต๋๋ค. |
|
3. **์ ํ ์ต์
**: ์์ ๋ฒํผ์ ํด๋ฆญํ๋ฉด ํ๋กฌํํธ ์
๋ ฅ๋์ ํ๊ตญ์ด ๋ฌธ๊ตฌ๋ก ์
๋ ฅํ์๋ฉด ๋ฉ๋๋ค. |
|
4. **๋ค์ํ ์ด๋ฏธ์ง**: "์ด๋ฏธ์ง ์์ฑ" ๋ฒํผ์ ํด๋ฆญํ๋ฉด ์ฐจ๋ก๋ก 4์ฅ์ ์ด๋ฏธ์ง๊ฐ ์์ฑ๋ฉ๋๋ค. |
|
5. **์์ ์ ํ**: ๋ค์ํ ์์ ๋ฅผ ํตํด ๋ฏธ๋ฆฌ ํ
์คํธํด๋ณด์ธ์. |
|
|
|
> **ํ**: ํ๋กฌํํธ๋ฅผ ์ง์ ์์ ํ ์๋ ์์ต๋๋ค. |
|
""" |
|
) |
|
|
|
|
|
gr.Markdown("## ์์ ์ด๋ฏธ์ง") |
|
examples = [ |
|
["down/๋ชจ๋ธ.jpg", None, None, "(#1์ ์ฌ์ฑ)์ด ์ด์ง ๋ค๋ก ๋์๋ณด๋ ๋ชจ์ต์ผ๋ก ์ต๋ํ ์ด์ seed๋ฅผ ์ ์งํํ
์์ฐ์ค๋ฝ๊ฒ ๋ณ๊ฒฝํ๋ผ."], |
|
["down/์์ด๋ ๊ณ ๋ชจํ.png", None, None, "(#1 ๋ ๋ชจ๋ชจํ)์์ ์ฒญ์์์ด๋ ๊ณ ๋ง ๊ฒ์์ ๊ณ ๋๋ ๊ณ ๋ก ๋ณ๊ฒฝํ๊ณ ๋๋จธ์ง ๋ถ๋ถ์ seed๋ฅผ ๋ณ๊ฒฝ์ ํ์ง๋ง๋ผ."], |
|
["down/์ค๊ตญ์ด.png", None, None, "(#1 ์ด๋ฏธ์ง)์ ์๋ ์ค๊ตญ์ด๋ฅผ ๋ชจ๋ ์ ๊ฑฐํ๋ผ."], |
|
["down/ํ
์คํธ.webp", None, None, '(#1์ ํ
์คํธ)๋ฅผ ์คํ์ผ์ ์ ์งํ์ฒด ํ
์คํธ๋ง "Hello"๋ก ๋ฐ๊ฟ๋ผ'], |
|
["down/๋ชจ๋ธ.jpg", "down/์ ๊ธ๋ผ์ค.png", "down/์ฒญ๋ฐ์ง.png", "(#1์ ์ฌ์ฑ๋ชจ๋ธ)์ด ์ ์ฒด ๋น์จ๊ณผ ํฌ์ฆ๋ ์ ์นํ ์ฒด (#2์ ์ ๊ธ๋ผ์ค)์ (#3์ ์ฒญ๋ฐ์ง)๋ฅผ ์ง์ ๋ชจ๋ธ์ด ์ฐฉ์ฉํ๊ฒ ์ฒ๋ผ ์์ฐ์ค๋ฝ๊ฒ ์ด๋ฏธ์ง๋ฅผ ์์ฑํ๋ผ."], |
|
["down/๋ชจ๋ธ.jpg", "down/์ ๊ธ๋ผ์ค.png", "down/์นดํ์ ๊ฒฝ.png", "(#1์ ์ฌ์ฑ๋ชจ๋ธ)์ด ์ ์ฒด ๋น์จ๊ณผ ํฌ์ฆ๋ ์ ์นํ ์ฒด (#2์ ์ ๊ธ๋ผ์ค)๋ฅผ ์ง์ ๋ชจ๋ธ์ด ์ฐฉ์ฉํ ๊ฒ์ฒ๋ผ (#3์ ์ฅ์)์์ ์์์ ์์ ์๋ ์์ฐ์ค๋ฌ์ด ์ด๋ฏธ์ง๋ฅผ ์์ฑํ๋ผ."], |
|
["down/๋ชจ๋ธ.jpg", "down/์์ธ์.png", None, "(#1์ ์ฌ์ฑ๋ชจ๋ธ)์ด ์ ์ฒด ๋น์จ๊ณผ ํฌ์ฆ๋ ์ ์นํ ์ฒด (#2์ ์์ธ์)์ ์ฌ์ฑ๋ชจ๋ธ์ด ํ๋ณดํ ์์ธ์์ ๋๋ณด์ด๊ฒ ๋ค๊ณ ์๋ ์์ฐ์ค๋ฌ์ด ๋ชจ์ต์ผ๋ก ์ด๋ฏธ์ง๋ฅผ ์์ฑํ๋ผ."], |
|
["down/๋ชจ๋ธ.jpg", "down/์นดํ์ ๊ฒฝ.png", None, "(#1์ ์ฌ์ฑ๋ชจ๋ธ)์ด (#2 ์ด๋ฏธ์ง์ ๋ฐฐ๊ฒฝ)์ ์ฃผ์ ํผ์ฌ์ฒด๋ ๊ทธ๋๋ก ์ ์งํ์ฌ ์ด๋ฏธ์ง์ ๋ถ์๊ธฐ๊ฐ ์์ฐ์ค๋ฝ๊ฒ ์ด์ฐ๋ฌ์ง๋๋ก ์์ฑํ๋ผ."], |
|
["down/์์ด๋ ๊ณ ๋ชจํ.png", None, None, "(#1์ ๋ ๊ณ ๋ชจํ)์์ ์ฒญ์์์ด๋ ๊ณ ๋ฅผ ์ ๊ฑฐํ ํ, ๊ทธ ์๋ฆฌ๋ฅผ ์ฃผ๋ณ ๋ฐฐ๊ฒฝ๊ณผ ์์ฐ์ค๋ฝ๊ฒ ์ด์ฐ๋ฌ์ง๋๋ก ์ฑ์์ฃผ์ธ์. ๋จ, ์ด๋ฏธ์ง์ ๋ค๋ฅธ ๋ถ๋ถ์ ์ฃผ์ ์์๋ ๋์ผํ๊ฒ ์ ์งํด ํด์ผํ๋ค."] |
|
] |
|
|
|
gr.Examples( |
|
examples=examples, |
|
inputs=[image1_input, image2_input, image3_input, prompt_input], |
|
elem_id="examples-grid" |
|
) |
|
|
|
|
|
with gr.Column(scale=1): |
|
with gr.Row(): |
|
with gr.Column(): |
|
output_image1 = gr.Image(label="์์ฑ๋ ์ด๋ฏธ์ง #1", height=600, width=450) |
|
output_image3 = gr.Image(label="์์ฑ๋ ์ด๋ฏธ์ง #3", height=600, width=450) |
|
with gr.Column(): |
|
output_image2 = gr.Image(label="์์ฑ๋ ์ด๋ฏธ์ง #2", height=600, width=450) |
|
output_image4 = gr.Image(label="์์ฑ๋ ์ด๋ฏธ์ง #4", height=600, width=450) |
|
|
|
output_text = gr.Textbox(label="์ํ ๋ฉ์์ง", lines=4) |
|
prompt_display = gr.Textbox(label="์ฌ์ฉ๋ ํ๋กฌํํธ (์์ด)", visible=True, lines=4) |
|
|
|
|
|
image_change_btn1.click( |
|
fn=lambda: "(#1์ ์ฌ์ฑ)์ด ์ด์ง ๋ค๋ก ๋์๋ณด๋ ๋ชจ์ต์ผ๋ก ์ต๋ํ ์ด์ seed๋ฅผ ์ ์งํํ
์์ฐ์ค๋ฝ๊ฒ ๋ณ๊ฒฝํ๋ผ.", |
|
inputs=[], |
|
outputs=prompt_input |
|
) |
|
image_change_btn2.click( |
|
fn=lambda: "(#1 ๋ ๋ชจ๋ชจํ)์์ ์ฒญ์์์ด๋ ๊ณ ๋ง ๊ฒ์์ ๊ณ ๋๋ ๊ณ ๋ก ๋ณ๊ฒฝํ๊ณ ๋๋จธ์ง ๋ถ๋ถ์ seed๋ฅผ ๋ณ๊ฒฝ์ ํ์ง๋ง๋ผ.", |
|
inputs=[], |
|
outputs=prompt_input |
|
) |
|
text_remove_btn.click( |
|
fn=lambda: "(#1 ์ด๋ฏธ์ง)์ ์๋ ์ค๊ตญ์ด๋ฅผ ๋ชจ๋ ์ ๊ฑฐํ๋ผ.", |
|
inputs=[], |
|
outputs=prompt_input |
|
) |
|
text_change_btn.click( |
|
fn=lambda: "(#1์ ํ
์คํธ)๋ฅผ ์คํ์ผ์ ์ ์งํ์ฒด ํ
์คํธ๋ง \"Hello\"๋ก ๋ฐ๊ฟ๋ผ", |
|
inputs=[], |
|
outputs=prompt_input |
|
) |
|
clothes_change_btn1.click( |
|
fn=lambda: "(#1์ ์ฌ์ฑ๋ชจ๊ฒ)์ด ์ ์ฒด ๋น์จ๊ณผ ํฌ์ฆ๋ ์ ์นํ ์ฒด (#2์ ์ ๊ธ๋ผ์ค)์ (#3์ ์ฒญ๋ฐ์ง)๋ฅผ ์ง์ ๋ชจ๋ธ์ด ์ฐฉ์ฉํ๊ฒ ์ฒ๋ผ ์์ฐ์ค๋ฝ๊ฒ ์ด๋ฏธ์ง๋ฅผ ์์ฑํ๋ผ.", |
|
inputs=[], |
|
outputs=prompt_input |
|
) |
|
clothes_change_btn2.click( |
|
fn=lambda: "(#1์ ์ฌ์ฑ๋ชจ๋ธ)์ด ์ ์ฒด ๋น์จ๊ณผ ํฌ์ฆ๋ ์ ์นํ ์ฒด (#2์ ์ ๊ธ๋ผ์ค)๋ฅผ ์ง์ ๋ชจ๋ธ์ด ์ฐฉ์ฉํ ๊ฒ์ฒ๋ผ (#3์ ์ฅ์)์์ ์์์ ์์ ์๋ ์์ฐ์ค๋ฌ์ด ์ด๋ฏธ์ง๋ฅผ ์์ฑํ๋ผ.", |
|
inputs=[], |
|
outputs=prompt_input |
|
) |
|
holding_product_btn.click( |
|
fn=lambda: "(#1์ ์ฌ์ฑ๋ชจ๋ธ)์ด ์ ์ฒด ๋น์จ๊ณผ ํฌ์ฆ๋ ์ ์นํ ์ฒด (#2์ ์์ธ์)์ ์ฌ์ฑ๋ชจ๋ธ์ด ํ๋ณดํ ์์ธ์์ ๋๋ณด์ด๊ฒ ๋ค๊ณ ์๋ ์์ฐ์ค๋ฌ์ด ๋ชจ์ต์ผ๋ก ์ด๋ฏธ์ง๋ฅผ ์์ฑํ๋ผ.", |
|
inputs=[], |
|
outputs=prompt_input |
|
) |
|
background_change_btn.click( |
|
fn=lambda: "(#1์ ์ฌ์ฑ๋ชจ๋ธ)์ด (#2 ์ด๋ฏธ์ง์ ๋ฐฐ๊ฒฝ)์ ์ฃผ์ ํผ์ฌ์ฒด๋ ๊ทธ๋๋ก ์ ์งํ์ฌ ๋ ์ด๋ฏธ์ง์ ๋ถ์๊ธฐ๊ฐ ์์ฐ์ค๋ฝ๊ฒ ์ด์ฐ๋ฌ์ง๋๋ก ์์ฑํ๋ผ.", |
|
inputs=[], |
|
outputs=prompt_input |
|
) |
|
composite_product_btn.click( |
|
fn=lambda: "(#1์ ๋ ๊ณ ๋ชจํ)์์ ์ฒญ์์์ด๋ ๊ณ ๋ฅผ ์ ๊ฑฐํ ํ, ๊ทธ ์๋ฆฌ๋ฅผ ์ฃผ๋ณ ๋ฐฐ๊ฒฝ๊ณผ ์์ฐ์ค๋ฝ๊ฒ ์ด์ฐ๋ฌ์ง๋๋ก ์ฑ์์ฃผ์ธ์. ๋จ, ์ด๋ฏธ์ง์ ๋ค๋ฅธ ๋ถ๋ถ์ ์ฃผ์ ์์๋ ๋์ผํ๊ฒ ์ ์งํด ํด์ผํ๋ค.", |
|
inputs=[], |
|
outputs=prompt_input |
|
) |
|
|
|
submit_btn.click( |
|
fn=generate_multiple_images, |
|
inputs=[image1_input, image2_input, image3_input, prompt_input], |
|
outputs=[output_image1, output_image2, output_image3, output_image4, output_text, prompt_display], |
|
) |
|
|
|
|
|
with gr.Tab("์ด๋ฏธ์ง ์
์ค์ผ์ผ๋ฌ"): |
|
|
|
download_upscaler_models() |
|
|
|
gr.Markdown( |
|
""" |
|
# ์ด๋ฏธ์ง ์
์ค์ผ์ผ๋ฌ |
|
|
|
์
๋ก๋ํ ์ด๋ฏธ์ง๋ฅผ ๊ณ ํด์๋๋ก ๋ณํํฉ๋๋ค. ํนํ ์ธ๋ฌผ ์ฌ์ง์ ์ผ๊ตด ๋ํ
์ผ์ ํฅ์์ํต๋๋ค. |
|
""" |
|
) |
|
|
|
with gr.Row(): |
|
with gr.Column(): |
|
input_upscale_image = gr.Image(type="filepath", label="์
๋ ฅ ์ด๋ฏธ์ง") |
|
upscale_btn = gr.Button("์ด๋ฏธ์ง ์
์ค์ผ์ผ", variant="primary") |
|
|
|
with gr.Column(): |
|
output_upscale_image = gr.Image(type="numpy", label="๊ฒฐ๊ณผ ์ด๋ฏธ์ง") |
|
|
|
upscale_btn.click( |
|
fn=upscale_image, |
|
inputs=[input_upscale_image], |
|
outputs=[output_upscale_image] |
|
) |
|
|
|
if __name__ == "__main__": |
|
demo.launch(share=True) |