server1.py

# comentarios sin tildes / sin enye

import io
import numpy as np
from fastapi import FastAPI, UploadFile, File
from fastapi.responses import JSONResponse
from PIL import Image
import torch
from transformers import SamModel, SamProcessor

app = FastAPI(title="Accudoctor Strip Analyzer")

DEVICE = "cuda" if torch.cuda.is_available() else "cpu"

model = SamModel.from_pretrained("facebook/sam-vit-base").to(DEVICE)
processor = SamProcessor.from_pretrained("facebook/sam-vit-base")


def detect_strip_mask(img):
    inputs = processor(img, return_tensors="pt").to(DEVICE)
    with torch.no_grad():
        out = model(**inputs)

    masks = processor.post_process_masks(
        out.pred_masks.cpu(),
        inputs["original_sizes"].cpu(),
        inputs["reshaped_input_sizes"].cpu()
    )[0].numpy()

    # cojo la mascara mas grande (la tira)
    best_mask = max(masks, key=lambda m: np.sum(m))

    best_mask = np.squeeze(best_mask)
    ys, xs = np.where(best_mask > 0.5)

    x1, x2 = xs.min(), xs.max()
    y1, y2 = ys.min(), ys.max()

    return x1, y1, x2, y2


def split_into_11(img_strip):
    w, h = img_strip.size
    block_h = h // 11

    blocks = []

    for i in range(11):
        y1 = i * block_h
        y2 = (i + 1) * block_h
        crop = img_strip.crop((0, y1, w, y2))
        blocks.append({
            "index": i + 1,
            "bbox": [0, y1, w, y2]
        })

    return blocks


def detect_blocks(image_bytes):

    img = Image.open(io.BytesIO(image_bytes)).convert("RGB")

    # 1) SAM detecta la tira completa
    x1, y1, x2, y2 = detect_strip_mask(img)

    strip = img.crop((x1, y1, x2, y2))

    # 2) se divide en 11 bloques
    blocks = split_into_11(strip)

    return blocks


@app.post("/strip/")
async def strip(front: UploadFile = File(...)):
    try:
        img_bytes = await front.read()
        blocks = detect_blocks(img_bytes)

        return JSONResponse(content={
            "code": 200,
            "num_blocks": len(blocks),
            "blocks": blocks
        })

    except Exception as e:
        return JSONResponse(content={"code": 500, "error": str(e)})