app.py

from fastapi import FastAPI, File, UploadFile
from fastapi.responses import JSONResponse
from fastapi.middleware.cors import CORSMiddleware
from PIL import Image
import io
import torch
from transformers import AutoImageProcessor, AutoBackbone
import pytesseract  # OCR

app = FastAPI()

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_methods=["*"],
    allow_headers=["*"],
)

processor = AutoImageProcessor.from_pretrained("czczup/textnet-base")
model = AutoBackbone.from_pretrained("czczup/textnet-base")
model.eval()

@app.post("/detect")
async def detect_text(file: UploadFile = File(...)):
    try:
        # Lire image
        image_bytes = await file.read()
        image = Image.open(io.BytesIO(image_bytes)).convert("RGB")

        # Entrée TextNet
        inputs = processor(image, return_tensors="pt")
        with torch.no_grad():
            outputs = model(**inputs)

        # Feature map et heatmap
        fm = outputs.feature_maps[-1][0]  # dernière layer
        heatmap = fm.mean(dim=0).numpy()
        H, W = heatmap.shape
        threshold = heatmap.max() * 0.2

        # Points chauds
        points = [(x, y) for y in range(H) for x in range(W) if heatmap[y, x] > threshold]

        if not points:
            return JSONResponse([])

        # Regrouper par lignes simples
        lines = {}
        for x, y in points:
            key = int(y / 10)
            lines.setdefault(key, []).append((x, y))

        # Générer boxes et extraire texte OCR
        scale_x = image.width / W
        scale_y = image.height / H
        boxes = []

        for line in lines.values():
            xs = [p[0] for p in line]
            ys = [p[1] for p in line]
            min_x, max_x = min(xs), max(xs)
            min_y, max_y = min(ys), max(ys)

            if (max_x - min_x) < 5 or (max_y - min_y) < 2:
                continue

            # Crop pour OCR
            crop = image.crop((
                int(min_x * scale_x),
                int(min_y * scale_y),
                int(max_x * scale_x),
                int(max_y * scale_y)
            ))

            text = pytesseract.image_to_string(crop, lang='eng').strip()

            if len(text) < 2:
                continue

            if len(boxes) == 0:
                boxes.append({
                    "x": 10,
                    "y": 10,
                    "w": 100,
                    "h": 50,
                    "text": "Aucun texte détecté"
                })

            boxes.append({
                "x": int(min_x * scale_x),
                "y": int(min_y * scale_y),
                "w": int((max_x - min_x) * scale_x),
                "h": int((max_y - min_y) * scale_y),
                "text": text or "texte non reconnu"
            })
            print("BOXES:", boxes)
    
        return JSONResponse({
            "image_width": image.width,
            "image_height": image.height,
            "boxes": boxes
        })

    except Exception as e:
        return JSONResponse({"success": False, "error": str(e)}, status_code=500)