import base64 import io import sys from pathlib import Path import numpy as np import torch from flask import Flask, jsonify, render_template, request from PIL import Image sys.path.insert(0, str(Path(__file__).resolve().parent.parent / "src")) from model import CNN # noqa: E402 from preprocess import preprocess # noqa: E402 app = Flask(__name__) DEVICE = "cuda" if torch.cuda.is_available() else "cpu" _model = CNN(depth=5, use_residual=False).to(DEVICE) _model.load_state_dict(torch.load( Path(__file__).resolve().parent.parent / "runs/depth5/model.pt", map_location=DEVICE, weights_only=True)) _model.eval() @app.route("/") def index(): return render_template("index.html") @app.route("/predict", methods=["POST"]) def predict(): data = request.json.get("image", "") png = base64.b64decode(data.split(",", 1)[-1]) img = Image.open(io.BytesIO(png)).convert("L") arr = np.array(img, dtype=np.float32) / 255.0 arr = preprocess(arr) with torch.no_grad(): x = torch.from_numpy(arr)[None, None].to(DEVICE) probs = torch.softmax(_model(x), dim=1)[0].cpu().numpy() top = probs.argsort()[::-1][:3] return jsonify({ "digit": int(top[0]), "confidence": float(probs[top[0]]), "top3": [{"digit": int(d), "prob": float(probs[d])} for d in top], }) if __name__ == "__main__": app.run(host="127.0.0.1", port=5000, debug=False)