app.py

import os
import gc
from abc import ABC, abstractmethod
from pathlib import Path
from typing import List, Dict, Any, Type

import cv2
import gradio as gr
import numpy as np
import pandas as pd
import torch
import onnxruntime as rt
from PIL import Image
from huggingface_hub import hf_hub_download
from transformers import pipeline, Pipeline
from tqdm import tqdm

# Suppress a specific PIL warning about image size
Image.MAX_IMAGE_PIXELS = None

# --- Configuration ---
CACHE_DIR = "./hf_cache"
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
DTYPE = torch.bfloat16 if torch.cuda.is_available() and torch.cuda.get_device_capability()[0] >= 8 else torch.float32

print(f"Using device: {DEVICE} with dtype: {DTYPE}")

# ==================================================================================
# 1. MODEL ABSTRACTION: A unified interface for all scorers.
# ==================================================================================

class AestheticScorer(ABC):
    """Abstract base class for all aesthetic scoring models."""

    def __init__(self, model_name: str, repo_id: str, filename: str = None):
        self.model_name = model_name
        self.repo_id = repo_id
        self.filename = filename
        self._model = None
        print(f"Initializing scorer: {self.model_name}")

    @property
    def model(self):
        """Lazy-loads the model on first access."""
        if self._model is None:
            print(f"Loading model for '{self.model_name}'...")
            self._model = self.load_model()
            print(f"'{self.model_name}' model loaded.")
        return self._model

    def _download_model(self) -> str:
        """Downloads the model file from Hugging Face Hub."""
        return hf_hub_download(repo_id=self.repo_id, filename=self.filename, cache_dir=CACHE_DIR)

    @abstractmethod
    def load_model(self) -> Any:
        """Loads the model and any necessary preprocessors."""
        pass

    @abstractmethod
    def score_batch(self, image_batch: List[Image.Image]) -> List[float]:
        """Scores a batch of images and returns a list of floats."""
        pass

    def release_model(self):
        """Releases model from memory."""
        if self._model is not None:
            print(f"Releasing model: {self.model_name}")
            del self._model
            self._model = None
            gc.collect()
            if torch.cuda.is_available():
                torch.cuda.empty_cache()

class PipelineScorer(AestheticScorer):
    """Scorer for models compatible with Hugging Face pipelines."""

    def load_model(self) -> Pipeline:
        """Loads a pipeline model."""
        return pipeline(
            "image-classification",
            model=self.repo_id,
            device=DEVICE,
        )

    @torch.no_grad()
    def score_batch(self, image_batch: List[Image.Image]) -> List[float]:
        """Scores a batch using the pipeline and extracts the 'hq' score."""
        results = self.model(image_batch)
        scores = []
        for res in results:
            try:
                # Find the score for the 'hq' (high quality) label
                hq_score = next(item['score'] for item in res if item['label'] == 'hq')
                scores.append(round(hq_score * 10.0, 4))
            except (StopIteration, TypeError):
                scores.append(0.0)
        return scores

class ONNXScorer(AestheticScorer):
    """Scorer for ONNX-based models."""

    def load_model(self) -> rt.InferenceSession:
        """Loads an ONNX inference session."""
        model_path = self._download_model()
        return rt.InferenceSession(model_path, providers=['CUDAExecutionProvider' if DEVICE == 'cuda' else 'CPUExecutionProvider'])

    def _preprocess(self, img: Image.Image) -> np.ndarray:
        """Preprocesses a single image for the Anime Aesthetic model."""
        img_np = np.array(img.convert("RGB")).astype(np.float32) / 255.0
        s = 768
        h, w = img_np.shape[:2]
        if h > w:
            new_h, new_w = s, int(s * w / h)
        else:
            new_h, new_w = int(s * h / w), s
        
        resized = cv2.resize(img_np, (new_w, new_h), interpolation=cv2.INTER_AREA)
        canvas = np.zeros((s, s, 3), dtype=np.float32)
        pad_h, pad_w = (s - new_h) // 2, (s - new_w) // 2
        canvas[pad_h:pad_h + new_h, pad_w:pad_w + new_w] = resized
        
        return np.transpose(canvas, (2, 0, 1))[np.newaxis, :]

    def score_batch(self, image_batch: List[Image.Image]) -> List[float]:
        """Scores images one by one as this model doesn't support batching."""
        scores = []
        for img in image_batch:
            try:
                input_tensor = self._preprocess(img)
                pred = self.model.run(None, {"img": input_tensor})[0].item()
                scores.append(round(pred * 10.0, 4))
            except Exception:
                scores.append(0.0)
        return scores

class CLIPMLPScorer(AestheticScorer):
    """Scorer for models using a CLIP backbone and an MLP head."""

    class MLP(torch.nn.Module):
        def __init__(self, input_size: int):
            super().__init__()
            self.layers = torch.nn.Sequential(
                torch.nn.Linear(input_size, 1024),
                torch.nn.ReLU(),
                torch.nn.Dropout(0.2),
                torch.nn.Linear(1024, 128),
                torch.nn.ReLU(),
                torch.nn.Dropout(0.2),
                torch.nn.Linear(128, 64),
                torch.nn.ReLU(),
                torch.nn.Linear(64, 16),
                torch.nn.ReLU(),
                torch.nn.Linear(16, 1),
            )
        def forward(self, x):
            return self.layers(x)

    def load_model(self) -> Dict[str, Any]:
        """Loads both the CLIP model and the custom MLP head."""
        import clip # Lazy import
        
        model_path = self._download_model()
        
        mlp = self.MLP(input_size=768) # ViT-L/14 has 768 features
        state_dict = torch.load(model_path, map_location=DEVICE)
        mlp.load_state_dict(state_dict)
        mlp.to(device=DEVICE, dtype=DTYPE)
        mlp.eval()

        clip_model, preprocess = clip.load("ViT-L/14", device=DEVICE)
        
        return {"mlp": mlp, "clip": clip_model, "preprocess": preprocess}

    @torch.no_grad()
    def score_batch(self, image_batch: List[Image.Image]) -> List[float]:
        """Scores a batch using CLIP features and the MLP head."""
        preprocess = self.model['preprocess']
        image_tensors = torch.cat([preprocess(img).unsqueeze(0) for img in image_batch]).to(DEVICE)
        
        image_features = self.model['clip'].encode_image(image_tensors)
        image_features /= image_features.norm(dim=-1, keepdim=True)
        
        # Pass features through MLP
        predictions = self.model['mlp'](image_features.to(DTYPE)).squeeze(-1)
        scores = predictions.float().cpu().numpy()
        
        return [round(float(s), 4) for s in scores]

# --- Model Registry ---
MODEL_REGISTRY: Dict[str, Type[AestheticScorer]] = {
    "Aesthetic Shadow V2": PipelineScorer(
        "Aesthetic Shadow V2", "shadowlilac/aesthetic-shadow-v2"
    ),
    "Waifu Scorer V2": CLIPMLPScorer(
        "Waifu Scorer V2", "skytnt/waifu-aesthetic-scorer", "model.pth"
    ),
    "Anime Scorer": ONNXScorer(
        "Anime Scorer", "skytnt/anime-aesthetic", "model.onnx"
    )
}

# In-memory cache for loaded model instances
_loaded_models_cache: Dict[str, AestheticScorer] = {}


# ==================================================================================
# 2. CORE PROCESSING LOGIC
# ==================================================================================

def get_scorers(model_names: List[str]) -> List[AestheticScorer]:
    """Retrieves and caches scorer instances based on selected names."""
    # Release models that are no longer selected
    for name, scorer in list(_loaded_models_cache.items()):
        if name not in model_names:
            scorer.release_model()
            del _loaded_models_cache[name]

    # Load newly selected models
    scorers = []
    for name in model_names:
        if name in _loaded_models_cache:
            scorers.append(_loaded_models_cache[name])
        elif name in MODEL_REGISTRY:
            scorer = MODEL_REGISTRY[name]
            _loaded_models_cache[name] = scorer
            scorers.append(scorer)
    return scorers
    
def evaluate_images(
    files: List[gr.File],
    selected_model_names: List[str],
    batch_size: int,
    progress: gr.Progress = gr.Progress(track_tqdm=True),
) -> pd.DataFrame:
    """
    Main function to process images, run them through selected models,
    and return results as a Pandas DataFrame.
    """
    if not files:
        gr.Warning("No images uploaded. Please upload files to evaluate.")
        return pd.DataFrame()
        
    if not selected_model_names:
        gr.Warning("No models selected. Please select at least one model.")
        return pd.DataFrame()

    try:
        image_paths = [Path(f.name) for f in files]
        all_results = []
        scorers = get_scorers(selected_model_names)
        
        # Use a single tqdm instance for progress tracking
        pbar = tqdm(total=len(image_paths), desc="Processing images")
        
        for i in range(0, len(image_paths), batch_size):
            batch_paths = image_paths[i : i + batch_size]
            
            # Load images for the current batch
            try:
                batch_images = [Image.open(p).convert("RGB") for p in batch_paths]
            except Exception as e:
                gr.Warning(f"Skipping a batch due to an error loading an image: {e}")
                pbar.update(len(batch_paths))
                continue

            # Get scores from all selected models for the batch
            batch_scores: Dict[str, List[float]] = {}
            for scorer in scorers:
                batch_scores[scorer.model_name] = scorer.score_batch(batch_images)
            
            # Collate results for the batch
            for j, path in enumerate(batch_paths):
                result_row = {"Image": Image.open(path), "Filename": path.name}
                
                scores_for_avg = []
                for scorer in scorers:
                    score = batch_scores[scorer.model_name][j]
                    result_row[scorer.model_name] = score
                    scores_for_avg.append(score)
                
                # Calculate average score
                if scores_for_avg:
                    result_row["Average Score"] = round(np.mean(scores_for_avg), 4)
                else:
                    result_row["Average Score"] = 0.0
                
                all_results.append(result_row)
            
            pbar.update(len(batch_paths))

        pbar.close()
        
        if not all_results:
            gr.Warning("Processing completed, but no results were generated.")
            return pd.DataFrame()
            
        return pd.DataFrame(all_results)
        
    except Exception as e:
        gr.Error(f"A critical error occurred: {e}")
        # Clean up in case of failure
        for scorer in _loaded_models_cache.values():
            scorer.release_model()
        _loaded_models_cache.clear()
        return pd.DataFrame()


# ==================================================================================
# 3. GRADIO USER INTERFACE
# ==================================================================================

def create_ui() -> gr.Blocks:
    """Creates and configures the Gradio web interface."""
    
    all_model_names = list(MODEL_REGISTRY.keys())
    
    # Define headers and datatypes for the results table
    dataframe_headers = ["Image", "Filename"] + all_model_names + ["Average Score"]
    dataframe_datatypes = ["image", "str"] + ["number"] * (len(all_model_names) + 1)
    
    with gr.Blocks(theme=gr.themes.Soft(), title="Image Aesthetic Scorer") as demo:
        gr.Markdown(
            """
            # 🖼️ Modern Image Aesthetic Scorer
            Upload your images, select the scoring models, and click 'Evaluate'.
            The results table supports **interactive sorting** (click on headers) and can be **downloaded as a CSV**.
            """
        )

        with gr.Row():
            with gr.Column(scale=1):
                gr.Markdown("### ⚙️ Settings")
                input_files = gr.Files(
                    label="Upload Images",
                    file_count="multiple",
                    file_types=["image"],
                )
                
                with gr.Accordion("Advanced Configuration", open=False):
                    model_checkboxes = gr.CheckboxGroup(
                        choices=all_model_names,
                        value=all_model_names,
                        label="Scoring Models",
                        info="Choose which models to use for evaluation.",
                    )
                    batch_size_slider = gr.Slider(
                        minimum=1,
                        maximum=64,
                        value=8,
                        step=1,
                        label="Batch Size",
                        info="Adjust based on your VRAM. Higher is faster.",
                    )

                with gr.Row():
                    process_button = gr.Button("🚀 Evaluate Images", variant="primary")
                    clear_button = gr.Button("🧹 Clear All")
            
            with gr.Column(scale=3):
                gr.Markdown("### 📊 Results")
                results_dataframe = gr.DataFrame(
                    headers=dataframe_headers,
                    datatype=dataframe_datatypes,
                    label="Evaluation Scores",
                    interactive=True,
                    # Enable the download button directly on the component
                )
                # This is a cleaner way to show the download button
                results_dataframe.style(height=800, show_download_button=True)


        # --- Event Handlers ---
        process_button.click(
            fn=evaluate_images,
            inputs=[input_files, model_checkboxes, batch_size_slider],
            outputs=[results_dataframe],
            concurrency_limit=1 # Only one evaluation at a time
        )

        def clear_outputs():
            # Release all models from memory when clearing
            for scorer in _loaded_models_cache.values():
                scorer.release_model()
            _loaded_models_cache.clear()
            gr.Info("Cleared results and released models from memory.")
            # Return an empty DataFrame to clear the table
            return pd.DataFrame()

        clear_button.click(
            fn=clear_outputs,
            inputs=[],
            outputs=[results_dataframe],
        )
        
    return demo

# ==================================================================================
# 4. APPLICATION ENTRY POINT
# ==================================================================================

if __name__ == "__main__":
    # Ensure cache directory exists
    os.makedirs(CACHE_DIR, exist_ok=True)
    
    app = create_ui()
    app.queue().launch(share=False)