models/classifier.py

from transformers import CLIPProcessor, CLIPModel
import torch
from PIL import Image
import io

class AttributeClassifier:
    def __init__(self):
        self.model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32")
        self.processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
        self.categories = ["clothing", "footwear", "furniture", "electronics", "accessories", "general"]
        # Attribute lists per category
        self.attributes = {
            "clothing": ["short sleeve", "long sleeve", "v-neck", "round neck", "solid", "striped", "button", "zipper"],
            "furniture": ["wood", "metal", "glass", "fabric", "modern", "vintage", "minimalist"],
            "footwear": ["sneaker", "boot", "loafer", "leather", "canvas"],
            "electronics": ["smartphone", "laptop", "tablet", "black", "silver"],
            "accessories": ["bag", "watch", "belt", "leather", "metal"]
        }
    
    def predict(self, image_bytes):
        image = Image.open(io.BytesIO(image_bytes)).convert('RGB')
        # First predict category
        cat_inputs = self.processor(text=self.categories, images=image, return_tensors="pt", padding=True)
        with torch.no_grad():
            cat_outputs = self.model(**cat_inputs)
        cat_logits = cat_outputs.logits_per_image[0]
        cat_probs = torch.softmax(cat_logits, dim=-1)
        best_cat_idx = cat_probs.argmax().item()
        best_category = self.categories[best_cat_idx]
        category_confidence = cat_probs[best_cat_idx].item()
        
        # Predict attributes for that category (if any)
        attr_list = self.attributes.get(best_category, [])
        attr_scores = {}
        if attr_list:
            attr_inputs = self.processor(text=attr_list, images=image, return_tensors="pt", padding=True)
            with torch.no_grad():
                attr_outputs = self.model(**attr_inputs)
            attr_logits = attr_outputs.logits_per_image[0]
            attr_probs = torch.softmax(attr_logits, dim=-1)
            attr_scores = {attr: prob.item() for attr, prob in zip(attr_list, attr_probs)}
        
        return {
            "category": best_category,
            "category_confidence": category_confidence,
            "attributes": attr_scores,
            "all_category_probs": {cat: prob.item() for cat, prob in zip(self.categories, cat_probs)}
        }

# Singleton
_classifier = None
def get_classifier():
    global _classifier
    if _classifier is None:
        _classifier = AttributeClassifier()
    return _classifier