app.py

import os
import shutil
import sys
import warnings
import random
import time
import logging
import fal_client
import base64
import numpy as np
import math
import scipy
import requests
import torch
import torchvision
import gradio as gr
import argparse
import spaces
from PIL import Image, ImageFilter, ImageOps, ImageDraw, ImageFont
from io import BytesIO
from typing import Dict, List, Tuple, Union, Optional

# Configure logging
logging.basicConfig(
    level=logging.INFO,
    format='%(asctime)s - %(name)s - %(levelname)s - %(message)s',
    handlers=[logging.StreamHandler()]
)
logger = logging.getLogger(__name__)

# Download model weights only if they don't exist
if not os.path.exists("groundingdino_swint_ogc.pth"):
    os.system("wget https://huggingface.co/ShilongLiu/GroundingDINO/resolve/main/groundingdino_swint_ogc.pth")

if not os.path.exists("sam_hq_vit_l.pth"):
    os.system("wget https://huggingface.co/lkeab/hq-sam/resolve/main/sam_hq_vit_l.pth")

# Add paths
sys.path.append(os.path.join(os.getcwd(), "GroundingDINO"))
sys.path.append(os.path.join(os.getcwd(), "sam-hq"))
warnings.filterwarnings("ignore")

# Grounding DINO
import GroundingDINO.groundingdino.datasets.transforms as T
from GroundingDINO.groundingdino.models import build_model
from GroundingDINO.groundingdino.util.slconfig import SLConfig
from GroundingDINO.groundingdino.util.utils import clean_state_dict, get_phrases_from_posmap

# segment anything
from segment_anything import build_sam_vit_l, SamPredictor 

# Constants
CONFIG_FILE = 'GroundingDINO/groundingdino/config/GroundingDINO_SwinT_OGC.py'
GROUNDINGDINO_CHECKPOINT = "groundingdino_swint_ogc.pth"
SAM_CHECKPOINT = 'sam_hq_vit_l.pth'
OUTPUT_DIR = "outputs"

# Global variables for model caching
_models = {
    'groundingdino': None,
    'sam_predictor': None
}

# Enable GPU if available with proper error handling
try:
    device = 'cuda' if torch.cuda.is_available() else 'cpu'
    logger.info(f"Using device: {device}")
except Exception as e:
    logger.warning(f"Error detecting GPU, falling back to CPU: {e}")
    device = 'cpu'


class ModelManager:
    """Manages model loading, unloading, and provides error handling"""
    
    @staticmethod
    def load_model(model_name: str) -> None:
        """Load a model if not already loaded"""
        try:
            if model_name == 'groundingdino' and _models['groundingdino'] is None:
                logger.info("Loading GroundingDINO model...")
                start_time = time.time()
                
                if not os.path.exists(GROUNDINGDINO_CHECKPOINT):
                    raise FileNotFoundError(f"GroundingDINO checkpoint not found at {GROUNDINGDINO_CHECKPOINT}")
                
                args = SLConfig.fromfile(CONFIG_FILE)
                args.device = device
                model = build_model(args)
                checkpoint = torch.load(GROUNDINGDINO_CHECKPOINT, map_location="cpu")
                load_res = model.load_state_dict(clean_state_dict(checkpoint["model"]), strict=False)
                logger.info(f"GroundingDINO load result: {load_res}")
                _ = model.eval()
                _models['groundingdino'] = model
                
                logger.info(f"GroundingDINO model loaded in {time.time() - start_time:.2f} seconds")
                
            elif model_name == 'sam' and _models['sam_predictor'] is None:
                logger.info("Loading SAM-HQ model...")
                start_time = time.time()
                
                if not os.path.exists(SAM_CHECKPOINT):
                    raise FileNotFoundError(f"SAM checkpoint not found at {SAM_CHECKPOINT}")
                
                sam = build_sam_vit_l(checkpoint=SAM_CHECKPOINT)
                sam.to(device=device)
                _models['sam_predictor'] = SamPredictor(sam)
                
                logger.info(f"SAM-HQ model loaded in {time.time() - start_time:.2f} seconds")

                
        except Exception as e:
            logger.error(f"Error loading {model_name} model: {e}")
            raise RuntimeError(f"Failed to load {model_name} model: {e}")
    
    @staticmethod
    def get_model(model_name: str):
        """Get a model, loading it if necessary"""
        if model_name not in _models or _models[model_name] is None:
            ModelManager.load_model(model_name)
        return _models[model_name]
    
    @staticmethod
    def unload_model(model_name: str) -> None:
        """Unload a model to free memory"""
        if model_name in _models and _models[model_name] is not None:
            logger.info(f"Unloading {model_name} model")
            _models[model_name] = None
            if device == 'cuda':
                torch.cuda.empty_cache()


def transform_image(image_pil: Image.Image) -> torch.Tensor:
    """Transform PIL image for GroundingDINO"""
    transform = T.Compose([
        T.RandomResize([800], max_size=1333),
        T.ToTensor(),
        T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
    ])
    image, _ = transform(image_pil, None)  # 3, h, w
    return image


def get_grounding_output(
    image: torch.Tensor, 
    caption: str, 
    box_threshold: float, 
    text_threshold: float, 
    with_logits: bool = True
) -> Tuple[torch.Tensor, torch.Tensor, List[str]]:
    """Run GroundingDINO to get bounding boxes from text prompt"""
    try:
        model = ModelManager.get_model('groundingdino')
        
        # Format caption
        caption = caption.lower().strip()
        if not caption.endswith("."):
            caption = caption + "."

        with torch.no_grad():
            outputs = model(image[None], captions=[caption])
            
        logits = outputs["pred_logits"].cpu().sigmoid()[0]  # (nq, 256)
        boxes = outputs["pred_boxes"].cpu()[0]  # (nq, 4)

        # Filter output
        logits_filt = logits.clone()
        boxes_filt = boxes.clone()
        filt_mask = logits_filt.max(dim=1)[0] > box_threshold
        logits_filt = logits_filt[filt_mask]  # num_filt, 256
        boxes_filt = boxes_filt[filt_mask]  # num_filt, 4

        # Get phrases
        tokenizer = model.tokenizer
        tokenized = tokenizer(caption)
        pred_phrases = []
        scores = []
        
        for logit, box in zip(logits_filt, boxes_filt):
            pred_phrase = get_phrases_from_posmap(
                logit > text_threshold, tokenized, tokenizer)
            if with_logits:
                pred_phrases.append(pred_phrase + f"({str(logit.max().item())[:4]})")
            else:
                pred_phrases.append(pred_phrase)
            scores.append(logit.max().item())

        return boxes_filt, torch.Tensor(scores), pred_phrases
    
    except Exception as e:
        logger.error(f"Error in grounding output: {e}")
        # Return empty results instead of crashing
        return torch.Tensor([]), torch.Tensor([]), []


def draw_mask(mask: np.ndarray, draw: ImageDraw.Draw) -> None:
    """Draw mask on image"""

    color = (255, 255, 255, 255)

    nonzero_coords = np.transpose(np.nonzero(mask))
    for coord in nonzero_coords:
        draw.point(coord[::-1], fill=color)


def draw_box(box: torch.Tensor, draw: ImageDraw.Draw, label: Optional[str]) -> None:
    """Draw bounding box on image"""
    color = tuple(np.random.randint(0, 255, size=3).tolist())
    draw.rectangle(((box[0], box[1]), (box[2], box[3])), outline=color, width=2)

    if label:
        font = ImageFont.load_default()
        if hasattr(font, "getbbox"):
            bbox = draw.textbbox((box[0], box[1]), str(label), font)
        else:
            w, h = draw.textsize(str(label), font)
            bbox = (box[0], box[1], w + box[0], box[1] + h)
        draw.rectangle(bbox, fill=color)
        draw.text((box[0], box[1]), str(label), fill="white")


def run_grounded_sam(input_image):
    """Main function to run GroundingDINO and SAM-HQ"""
    # Create output directory
    os.makedirs(OUTPUT_DIR, exist_ok=True)
    text_prompt = 'car'
    task_type = 'text'
    box_threshold = 0.3
    text_threshold = 0.25
    iou_threshold = 0.8
    hq_token_only = True
    
    # Process input image
    if isinstance(input_image, dict):
        # Input from gradio sketch component
        scribble = np.array(input_image["mask"])
        image_pil = input_image["image"].convert("RGB")
    else:
        # Direct image input
        image_pil = input_image.convert("RGB") if input_image else None
        scribble = None
        
    if image_pil is None:
        logger.error("No input image provided")
        return [Image.new('RGB', (400, 300), color='gray')]

    # Transform image for GroundingDINO
    transformed_image = transform_image(image_pil)
    
    # Load models as needed
    ModelManager.load_model('groundingdino')
    size = image_pil.size
    H, W = size[1], size[0]

    # Run GroundingDINO with provided text
    boxes_filt, scores, pred_phrases = get_grounding_output(
        transformed_image, text_prompt, box_threshold, text_threshold
    )

    if boxes_filt is not None:
        # Scale boxes to image dimensions
        for i in range(boxes_filt.size(0)):
            boxes_filt[i] = boxes_filt[i] * torch.Tensor([W, H, W, H])
            boxes_filt[i][:2] -= boxes_filt[i][2:] / 2
            boxes_filt[i][2:] += boxes_filt[i][:2]
        
        # Apply non-maximum suppression if we have multiple boxes
        if boxes_filt.size(0) > 1:
            logger.info(f"Before NMS: {boxes_filt.shape[0]} boxes")
            nms_idx = torchvision.ops.nms(boxes_filt, scores, iou_threshold).numpy().tolist()
            boxes_filt = boxes_filt[nms_idx]
            pred_phrases = [pred_phrases[idx] for idx in nms_idx]
            logger.info(f"After NMS: {boxes_filt.shape[0]} boxes")
    
    # Load SAM model
    ModelManager.load_model('sam')
    sam_predictor = ModelManager.get_model('sam_predictor')
    
    # Set image for SAM
    image = np.array(image_pil)
    sam_predictor.set_image(image)
    
    # Run SAM 
    # Use boxes for these task types
    if boxes_filt.size(0) == 0:
        logger.warning("No boxes detected")
        return [image_pil, Image.new('RGBA', size, color=(0, 0, 0, 0))]
    
    transformed_boxes = sam_predictor.transform.apply_boxes_torch(boxes_filt, image.shape[:2]).to(device)
    
    masks, _, _ = sam_predictor.predict_torch(
        point_coords=None,
        point_labels=None,
        boxes=transformed_boxes,
        multimask_output=False,
        hq_token_only=hq_token_only,
    )
    
    # Create mask image
    mask_image = Image.new('RGBA', size, color=(0, 0, 0, 0))
    mask_draw = ImageDraw.Draw(mask_image)
    
    # Draw masks
    for mask in masks:
        draw_mask(mask[0].cpu().numpy(), mask_draw)
    
    # Draw boxes and points on original image
    image_draw = ImageDraw.Draw(image_pil)

    for box, label in zip(boxes_filt, pred_phrases):
        draw_box(box, image_draw, label)

    return mask_image
    
    # except Exception as e:
    #     logger.error(f"Error in run_grounded_sam: {e}")
        # # Return original image on error
        # if isinstance(input_image, dict) and "image" in input_image:
        #     return [input_image["image"], Image.new('RGBA', input_image["image"].size, color=(0, 0, 0, 0))]
        # elif isinstance(input_image, Image.Image):
        #     return [input_image, Image.new('RGBA', input_image.size, color=(0, 0, 0, 0))]
        # else:
        #     return [Image.new('RGB', (400, 300), color='gray'), Image.new('RGBA', (400, 300), color=(0, 0, 0, 0))]

def split_image_with_alpha(image):
    image = image.convert("RGB")
    return image

def gaussian_blur(image, radius=10):
    """Apply Gaussian blur to image."""
    blurred = image.filter(ImageFilter.GaussianBlur(radius=10))
    return blurred

def invert_image(image):
    img_inverted = ImageOps.invert(image)
    return img_inverted

def expand_mask(mask, expand, tapered_corners):
    # Ensure mask is in grayscale (mode 'L')
    mask = mask.convert("L")
    
    # Convert to NumPy array
    mask_np = np.array(mask)

    # Define kernel
    c = 0 if tapered_corners else 1
    kernel = np.array([[c, 1, c],
                       [1, 1, 1],
                       [c, 1, c]], dtype=np.uint8)

    # Perform dilation or erosion based on expand value
    if expand > 0:
        for _ in range(expand):
            mask_np = scipy.ndimage.grey_dilation(mask_np, footprint=kernel)
    elif expand < 0:
        for _ in range(abs(expand)):
            mask_np = scipy.ndimage.grey_erosion(mask_np, footprint=kernel)

    # Convert back to PIL image
    return Image.fromarray(mask_np, mode="L")

def image_blend_by_mask(image_a, image_b, mask, blend_percentage):
    # Ensure images have the same size and mode
    image_a = image_a.convert('RGB')
    image_b = image_b.convert('RGB')
    mask = mask.convert('L')
    
    # Resize images if they don't match
    if image_a.size != image_b.size:
        image_b = image_b.resize(image_a.size, Image.LANCZOS)
    
    # Ensure mask has the same size
    if mask.size != image_a.size:
        mask = mask.resize(image_a.size, Image.LANCZOS)
    
    # Invert mask
    mask = ImageOps.invert(mask)

    # Mask image
    masked_img = Image.composite(image_a, image_b, mask)

    # Blend image
    blend_mask = Image.new(mode="L", size=image_a.size,
                           color=(round(blend_percentage * 255)))
    blend_mask = ImageOps.invert(blend_mask)
    img_result = Image.composite(image_a, masked_img, blend_mask)

    del image_a, image_b, blend_mask, mask

    return img_result

def blend_images(image_a, image_b, blend_percentage):
    """Blend img_b over image_a using the normal mode with a blend percentage."""
    img_a = image_a.convert("RGBA")
    img_b = image_b.convert("RGBA")

    # Blend img_b over img_a using alpha_composite (normal blend mode)
    out_image = Image.alpha_composite(img_a, img_b)

    out_image = out_image.convert("RGB")

    # Create blend mask
    blend_mask = Image.new("L", image_a.size, round(blend_percentage * 255))
    blend_mask = ImageOps.invert(blend_mask)  # Invert the mask

    # Apply composite blend
    result = Image.composite(image_a, out_image, blend_mask)
    return result

def apply_image_levels(image, black_level, mid_level, white_level):
    levels = AdjustLevels(black_level, mid_level, white_level)
    adjusted_image = levels.adjust(image)
    return adjusted_image

class AdjustLevels:
    def __init__(self, min_level, mid_level, max_level):
        self.min_level = min_level
        self.mid_level = mid_level
        self.max_level = max_level

    def adjust(self, im):

        im_arr = np.array(im).astype(np.float32)
        im_arr[im_arr < self.min_level] = self.min_level
        im_arr = (im_arr - self.min_level) * \
            (255 / (self.max_level - self.min_level))
        im_arr = np.clip(im_arr, 0, 255)

        # mid-level adjustment
        gamma = math.log(0.5) / math.log((self.mid_level - self.min_level) / (self.max_level - self.min_level))
        im_arr = np.power(im_arr / 255, gamma) * 255

        im_arr = im_arr.astype(np.uint8)

        im = Image.fromarray(im_arr)

        return im

def resize_image(image, scaling_factor=1):
    image = image.resize((int(image.width * scaling_factor), 
        int(image.height * scaling_factor)))
    return image

def resize_to_square(image, size=1024):

    # Load image if a file path is provided
    if isinstance(image, str):
        img = Image.open(image).convert("RGBA")
    else:
        img = image.convert("RGBA")  # If already an Image object

    # Resize while maintaining aspect ratio
    img.thumbnail((size, size), Image.LANCZOS)

    # Create a transparent square canvas
    square_img = Image.new("RGBA", (size, size), (0, 0, 0, 0))

    # Calculate the position to paste the resized image (centered)
    x_offset = (size - img.width) // 2
    y_offset = (size - img.height) // 2

    # Extract the alpha channel as a mask
    mask = img.split()[3] if img.mode == "RGBA" else None

    # Paste the resized image onto the square canvas with the correct transparency mask
    square_img.paste(img, (x_offset, y_offset), mask)

    return square_img


def encode_image(image):
    buffer = BytesIO()
    image.save(buffer, format="PNG")
    encoded_image = base64.b64encode(buffer.getvalue()).decode("utf-8")
    return f"data:image/png;base64,{encoded_image}"

def generate_ai_bg(input_img, prompt):
    # input_img = resize_image(input_img, 0.01)
    hf_input_img = encode_image(input_img)

    handler = fal_client.submit(
        "fal-ai/iclight-v2",
        arguments={
            "prompt": prompt,
            "image_url": hf_input_img
        },
        webhook_url="https://optional.webhook.url/for/results",
    )

    request_id = handler.request_id

    status = fal_client.status("fal-ai/iclight-v2", request_id, with_logs=True)

    result = fal_client.result("fal-ai/iclight-v2", request_id)

    relight_img_path = result['images'][0]['url']

    response = requests.get(relight_img_path, stream=True)

    relight_img = Image.open(BytesIO(response.content)).convert("RGBA")

    # from gradio_client import Client, handle_file

    # client = Client("lllyasviel/iclight-v2-vary")

    # result = client.predict(
    #         input_fg=handle_file(input_img),
    #         bg_source="None",
    #         prompt=prompt,
    #         image_width=256,
    #         image_height=256,
    #         num_samples=1,
    #         seed=12345,
    #         steps=25,
    #         n_prompt="lowres, bad anatomy, bad hands, cropped, worst quality",
    #         cfg=2,
    #         gs=5,
    #         enable_hr_fix=True,
    #         hr_downscale=0.5,
    #         lowres_denoise=0.8,
    #         highres_denoise=0.99,
    #         api_name="/process"
    # )
    # print(result)

    # relight_img_path = result[0][0]['image']

    # relight_img = Image.open(relight_img_path).convert("RGBA")

    return relight_img

def blend_details(input_image, relit_image, masked_image, scaling_factor=1):

    # input_image = resize_image(input_image)

    # relit_image = resize_image(relit_image)

    # masked_image = resize_image(masked_image)

    masked_image_rgb = split_image_with_alpha(masked_image)
    masked_image_blurred = gaussian_blur(masked_image_rgb, radius=10)
    grow_mask = expand_mask(masked_image_blurred, -15, True)

    # grow_mask.save("output/grow_mask.png")

    # Split images and get RGB channels
    input_image_rgb = split_image_with_alpha(input_image)
    input_blurred = gaussian_blur(input_image_rgb, radius=10)
    input_inverted = invert_image(input_image_rgb)

    # input_blurred.save("output/input_blurred.png")
    # input_inverted.save("output/input_inverted.png")
        
    # Add blurred and inverted images
    input_blend_1 = blend_images(input_inverted, input_blurred, blend_percentage=0.5)
    input_blend_1_inverted = invert_image(input_blend_1)
    input_blend_2 = blend_images(input_blurred, input_blend_1_inverted, blend_percentage=1.0)

    # input_blend_2.save("output/input_blend_2.png")

    # Process relit image
    relit_image_rgb = split_image_with_alpha(relit_image)
    relit_blurred = gaussian_blur(relit_image_rgb, radius=10)
    relit_inverted = invert_image(relit_image_rgb)

    # relit_blurred.save("output/relit_blurred.png")
    # relit_inverted.save("output/relit_inverted.png")
    
    # Add blurred and inverted relit images
    relit_blend_1 = blend_images(relit_inverted, relit_blurred, blend_percentage=0.5)
    relit_blend_1_inverted = invert_image(relit_blend_1)
    relit_blend_2 = blend_images(relit_blurred, relit_blend_1_inverted, blend_percentage=1.0)

    # relit_blend_2.save("output/relit_blend_2.png")

    high_freq_comp = image_blend_by_mask(relit_blend_2, input_blend_2, grow_mask, blend_percentage=1.0)

    # high_freq_comp.save("output/high_freq_comp.png")

    comped_image = blend_images(relit_blurred, high_freq_comp, blend_percentage=0.65)

    # comped_image.save("output/comped_image.png")

    final_image = apply_image_levels(comped_image, black_level=83, mid_level=128, white_level=172)

    # final_image.save("output/final_image.png")

    return final_image

@spaces.GPU
def generate_image(input_image_path, prompt):

    resized_input_img = resize_to_square(input_image_path, 256)

    resized_input_img_path = '/tmp/gradio/resized_input_img.png'

    resized_input_img.convert("RGBA").save(resized_input_img_path, "PNG")

    ai_gen_image = generate_ai_bg(resized_input_img, prompt)

    # mask_input_image = run_grounded_sam(resized_input_img)

    # final_image = blend_details(resized_input_img, ai_gen_image, mask_input_image)

    final_image = ai_gen_image

    return final_image

def create_ui():
    """Create Gradio UI for CarViz demo"""
    with gr.Blocks(title="CarViz Demo") as block:
        gr.Markdown("""
        # CarViz
        """)

        with gr.Row():
            with gr.Column():
                input_image_path = gr.Image(type="filepath", label="image")
                # ai_image = gr.Image(type="pil", label="image")
                prompt = gr.Textbox(label="Prompt", placeholder="Enter your prompt here...")
                run_button = gr.Button(value='Run')

            with gr.Column():
                output_image = gr.Image(label="Generated Image")
        
        # Run button
        run_button.click(
            fn=generate_image, 
            inputs=[
                input_image_path,
                # ai_image,
                prompt
            ], 
            outputs=[output_image]
        )     
 
    return block


if __name__ == "__main__":
    parser = argparse.ArgumentParser("Carviz demo", add_help=True)
    parser.add_argument("--debug", action="store_true", help="using debug mode")
    parser.add_argument("--share", action="store_true", help="share the app")
    parser.add_argument('--no-gradio-queue', action="store_true", help="disable gradio queue")
    parser.add_argument('--port', type=int, default=7860, help="port to run the app")
    parser.add_argument('--host', type=str, default="0.0.0.0", help="host to run the app")
    args = parser.parse_args()

    logger.info(f"Starting CarViz demo with args: {args}")
    
    # Check for model files
    if not os.path.exists(GROUNDINGDINO_CHECKPOINT):
        logger.warning(f"GroundingDINO checkpoint not found at {GROUNDINGDINO_CHECKPOINT}")
    if not os.path.exists(SAM_CHECKPOINT):
        logger.warning(f"SAM-HQ checkpoint not found at {SAM_CHECKPOINT}")
    
    # Create app
    block = create_ui()
    if not args.no_gradio_queue:
        block = block.queue()
    
    # Launch app
    try:
        block.launch(
            debug=args.debug, 
            share=args.share, 
            show_error=True,
            server_name=args.host,
            server_port=args.port
        )
    except Exception as e:
        logger.error(f"Error launching app: {e}")