# app.py import os import uuid from io import BytesIO from typing import Dict from fastapi import FastAPI, File, Form, UploadFile from fastapi.middleware.cors import CORSMiddleware from pydantic import BaseModel from PIL import Image from rembg import remove import torch import clip from supabase import create_client # ─── Supabase client setup ───────────────────────── SUPABASE_URL = os.environ["SUPABASE_URL"] SUPABASE_KEY = os.environ["SUPABASE_SERVICE_ROLE_KEY"] supabase = create_client(SUPABASE_URL, SUPABASE_KEY) # ─── CLIP zero-shot setup ────────────────────────── device, _ = ("cuda", None) if torch.cuda.is_available() else ("cpu", None) clip_model, clip_preprocess = clip.load("ViT-B/32", device=device) CATEGORIES = [ # Tops "t-shirt","dress shirt","blouse","tank top","sweater","hoodie","jacket","coat", # Bottoms "jeans","pants","trousers","chinos","leggings","shorts","skirt", # Footwear "sneakers","boots","sandals","heels","loafers","flats", # Under-garments "boxer briefs","bra","vest", # Accessories "ring","watch","bracelet","necklace","belt","hat", ] app = FastAPI() app.add_middleware( CORSMiddleware, allow_origins=["*"], allow_methods=["POST"], allow_headers=["*"], ) class Processed(BaseModel): image_url: str category: str def remove_bg_and_composite(img: Image.Image) -> Image.Image: fg = remove(img.convert("RGBA")) white = Image.new("RGBA", fg.size, (255,255,255,255)) white.paste(fg, mask=fg.split()[3]) return white.convert("RGB") def classify(img: Image.Image) -> str: img_t = clip_preprocess(img).unsqueeze(0).to(device) txt_t = clip.tokenize(CATEGORIES).to(device) with torch.no_grad(): feats_i = clip_model.encode_image(img_t) feats_t = clip_model.encode_text(txt_t) feats_i /= feats_i.norm(dim=-1, keepdim=True) feats_t /= feats_t.norm(dim=-1, keepdim=True) sims = (feats_i @ feats_t.T).squeeze(0) return CATEGORIES[sims.argmax().item()] @app.post("/process", response_model=Processed) async def process_and_store( user_id: str = Form(...), file: UploadFile = File(...) ) -> Dict[str,str]: # 1) load & remove bg content = await file.read() orig = Image.open(BytesIO(content)) processed = remove_bg_and_composite(orig) # 2) JPEG→bytes buf = BytesIO() processed.save(buf, format="JPEG", quality=95) buf.seek(0) # 3) upload to Supabase Storage filename = f"{user_id}/{uuid.uuid4().hex}.jpg" supabase.storage.from_("userwardrobe").upload( filename, buf.getvalue(), {"content-type": "image/jpeg"} ) # 4) get public URL (returns string) public_url = supabase.storage.from_("userwardrobe").get_public_url(filename) # 5) classify + insert row category = classify(processed) supabase.table("userwardrobe").insert({ "user_id": user_id, "image_url": public_url, "category": category }).execute() return {"image_url": public_url, "category": category}