app.py

import gradio as gr
import os
import pandas as pd
from PIL import Image
import torch
from transformers import BlipProcessor, BlipForConditionalGeneration
import shutil

# Load BLIP model
processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base")

# Keyword groups
shirt_keywords = ["shirt", "hoodie", "sweatshirt", "jacket", "tee", "top"]
bottom_keywords = ["pants", "shorts", "jeans", "skirt", "trousers", "bottom"]
boy_keywords = ["boy", "boys", "male", "mens"]
girl_keywords = ["girl", "girls", "female", "womens"]
color_keywords = [
    "black", "white", "red", "orange", "yellow", "green", "blue", "purple",
    "brown", "gray", "pink", "beige", "tan", "navy", "olive", "teal", "burgundy",
    "lavender", "peach", "mint", "charcoal", "gold", "silver"
]

def tag_and_sort_images(images):
    os.makedirs("uploaded_images", exist_ok=True)
    os.makedirs("sorted_images", exist_ok=True)

    results = []

    for image in images:
        image_path = os.path.join("uploaded_images", os.path.basename(image.name))
        shutil.copy2(image.name, image_path)

        raw_image = Image.open(image_path).convert('RGB')
        inputs = processor(raw_image, return_tensors="pt")
        out = model.generate(**inputs)
        caption = processor.decode(out[0], skip_special_tokens=True)

        results.append({
            "filename": os.path.basename(image.name),
            "caption": caption
        })

    # Build DataFrame
    df = pd.DataFrame(results)

    def detect_garment_type(text):
        t = text.lower()
        if any(k in t for k in shirt_keywords):
            return "Shirt"
        elif any(k in t for k in bottom_keywords):
            return "Bottom"
        else:
            return "Unknown"

    def detect_gender(text):
        t = text.lower()
        if any(k in t for k in boy_keywords):
            return "Boys"
        elif any(k in t for k in girl_keywords):
            return "Girls"
        else:
            return "Unisex"

    def extract_colors(text):
        return ', '.join(sorted(set(c for c in color_keywords if c in text.lower())))

    df["garment_type"] = df["caption"].apply(detect_garment_type)
    df["gender"] = df["caption"].apply(detect_gender)
    df["color_group"] = df["caption"].apply(extract_colors)

    df = df[["filename", "caption", "garment_type", "gender", "color_group"]]
    df.to_csv("blip2_image_tags_clean.csv", index=False)

    # Sorting
    for _, row in df.iterrows():
        filename = row["filename"]
        garment = row["garment_type"] or "Unsorted"
        gender = row["gender"] or "Unsorted"
        color = row["color_group"] or "Unknown"

        target_path = os.path.join("sorted_images", garment, gender, color)
        os.makedirs(target_path, exist_ok=True)

        src_path = os.path.join("uploaded_images", filename)
        dst_path = os.path.join(target_path, filename)

        if os.path.exists(src_path):
            shutil.copy2(src_path, dst_path)

    zip_path = shutil.make_archive("sorted_images", 'zip', "sorted_images")
    return zip_path

demo = gr.Interface(
    fn=tag_and_sort_images,
    inputs=gr.File(file_types=["image"], file_count="multiple", label="Upload JPG or PNG Images"),
    outputs=gr.File(label="Download Sorted ZIP"),
    title="🧠 Auto Tag & Sort Images",
    description="Upload fashion images. AI will tag them (garment, gender, color), sort them into folders, and give you a download link."
)

demo.launch()