enhancer (#1)

- enhancer (93eb6fd1c0e656edf92df419f1e94abee049610c)
- Update image_enhancer.oy (8182ca42a950217bedfb4c52cc1154b847b11325)
- Rename image_enhancer.oy to image_enhancer.py (1e45b793cbfd40f78975336beb12e506921a8b9d)
- Update image_enhancer.py (57a5562af893a0266d0c2cc72805ed3a9c9437e3)
- Update app.py (28eb981e002487de93d28a09ac42c0db99d52f22)
- Update app.py (d0a4bd9894d22e496eae32f1b8c4e1d468589c54)

app.py

@@ -1,7 +1,51 @@
-from fastapi import FastAPI
+from fastapi import FastAPI, File, UploadFile, HTTPException
+from fastapi.responses import StreamingResponse
+from media_enhancer.image_enhancer import EnhancementMethod, Enhancer
+from pydantic import BaseModel
+from PIL import Image
+from io import BytesIO
+import base64
+import numpy as np
+
+class EnhancementRequest(BaseModel):
+    method: EnhancementMethod = EnhancementMethod.gfpgan
+    background_enhancement: bool = True
+    upscale: int = 2
+
+class _EnhanceBase(BaseModel):
+    encoded_base_img: List[str]
+
+
 
 app = FastAPI()
 
 @app.get("/")
 def greet_json():
-    return {"Initializing GlamApp Enhancer"}
+    return {"Initializing GlamApp Enhancer"}
+
+@app.post("/enhance")
+async def enhance_image(
+    file: UploadFile = File(...),
+    request: EnhancementRequest = EnhancementRequest()
+):
+    try:
+        if not file.content_type.startswith('image/'):
+            raise HTTPException(status_code=400, detail="Invalid file type")
+            
+        contents = await file.read()
+        base64_encoded_image = base64.b64encode(contents).decode('utf-8')
+        data = _EnhanceBase(encoded_base_img=[base64_encoded_image])
+
+        enhancer = Enhancer(request.method, request.background_enhancement, request.upscale)
+
+        enhanced_img, original_resolution, enhanced_resolution = await enhancer.enhance(data)
+
+        enhanced_image = Image.fromarray(enhanced_img)
+        img_byte_arr = BytesIO()
+        enhanced_image.save(img_byte_arr, format='PNG')
+        img_byte_arr.seek(0)
+        
+        return StreamingResponse(img_byte_arr, media_type="image/png")
+
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))

image_enhancer.py

@@ -0,0 +1,123 @@
+import os
+import torch 
+from gfpgan import GFPGANer
+from tqdm import tqdm
+import cv2
+from realesrgan import RealESRGANer
+from basicsr.archs.rrdbnet_arch import RRDBNet
+import warnings
+from enum import Enum
+
+class EnhancementMethod(str, Enum):
+    gfpgan = "gfpgan"
+    RestoreFormer = "RestoreFormer"
+    codeformer = "codeformer"
+    realesrgan = "realesrgan"
+
+
+class Enhancer:
+    def __init__(self, method: EnhancementMethod, background_enhancement=True, upscale=2):
+        self.method = method
+        self.background_enhancement = background_enhancement
+        self.upscale = upscale
+        self.bg_upsampler = None
+        self.realesrgan_enhancer = None
+
+        if self.method != EnhancementMethod.realesrgan:
+            self.setup_face_enhancer()
+            if self.background_enhancement:
+                self.setup_background_enhancer()
+        else:
+            self.setup_realesrgan_enhancer()
+
+    def setup_background_enhancer(self):
+        if not torch.cuda.is_available():
+            warnings.warn('The unoptimized RealESRGAN is slow on CPU. We do not use it.')
+            return
+
+        model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=23, num_grow_ch=32, scale=self.upscale)
+        model_path = f'https://huggingface.co/dtarnow/UPscaler/resolve/main/RealESRGAN_x{self.upscale}plus.pth'
+        self.bg_upsampler = RealESRGANer(
+            scale=self.upscale,
+            model_path=model_path,
+            model=model,
+            tile=400,
+            tile_pad=10,
+            pre_pad=0,
+            half=True)
+
+    def setup_realesrgan_enhancer(self):
+        if not torch.cuda.is_available():
+            raise ValueError('CUDA is not available for RealESRGAN')
+
+        model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=23, num_grow_ch=32, scale=self.upscale)
+        model_path = f'https://huggingface.co/dtarnow/UPscaler/resolve/main/RealESRGAN_x{self.upscale}plus.pth'
+        self.realesrgan_enhancer = RealESRGANer(
+            scale=self.upscale,
+            model_path=model_path,
+            model=model,
+            tile=400,
+            tile_pad=10,
+            pre_pad=0,
+            half=True)
+
+    def setup_face_enhancer(self):
+        model_configs = {
+            EnhancementMethod.gfpgan: {
+                'arch': 'clean',
+                'channel_multiplier': 2,
+                'model_name': 'GFPGANv1.4',
+                'url': 'https://huggingface.co/gmk123/GFPGAN/resolve/main/GFPGANv1.4.pth'
+            },
+            EnhancementMethod.RestoreFormer: {
+                'arch': 'RestoreFormer',
+                'channel_multiplier': 2,
+                'model_name': 'RestoreFormer',
+                'url': 'https://github.com/TencentARC/GFPGAN/releases/download/v1.3.4/RestoreFormer.pth'
+            },
+            EnhancementMethod.codeformer: {
+                'arch': 'CodeFormer',
+                'channel_multiplier': 2,
+                'model_name': 'CodeFormer',
+                'url': 'https://huggingface.co/sinadi/aar/resolve/main/codeformer.pth'
+            }
+        }
+
+        config = model_configs.get(self.method)
+        if not config:
+            raise ValueError(f'Wrong model version {self.method}')
+
+        model_path = os.path.join('gfpgan/weights', config['model_name'] + '.pth')
+        if not os.path.isfile(model_path):
+            model_path = os.path.join('checkpoints', config['model_name'] + '.pth')
+        if not os.path.isfile(model_path):
+            model_path = config['url']
+
+        self.face_enhancer = GFPGANer(
+            model_path=model_path,
+            upscale=self.upscale,
+            arch=config['arch'],
+            channel_multiplier=config['channel_multiplier'],
+            bg_upsampler=self.bg_upsampler)
+
+    def check_image_resolution(self, image):
+        height, width, _ = image.shape
+        return width, height
+
+    async def enhance(self, image):
+        img = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
+        width, height = self.check_image_resolution(img)
+        
+        if self.method == EnhancementMethod.realesrgan:
+            enhanced_img, _ = await asyncio.to_thread(self.realesrgan_enhancer.enhance, img, outscale=self.upscale)
+        else:
+            _, _, enhanced_img = await asyncio.to_thread(self.face_enhancer.enhance,
+                img,
+                has_aligned=False,
+                only_center_face=False,
+                paste_back=True)
+        
+        enhanced_img = cv2.cvtColor(enhanced_img, cv2.COLOR_BGR2RGB)
+        enhanced_width, enhanced_height = self.check_image_resolution(enhanced_img)
+
+        return enhanced_img, (width, height), (enhanced_width, enhanced_height)