app.py

import os
import cv2
import json
import torch
import logging

import numpy as np
from flask import Flask, jsonify, request
from flask.wrappers import Response

from iam_line_recognition.model_main import CRNN
from iam_line_recognition.utils import ctc_decode
from iam_line_recognition.dataset import HWRecogIAMDataset

app = Flask(__name__)
logging.basicConfig(level=logging.INFO)

file_model_local = "artifacts/crnn_H_32_W_768_E_196.pth"
file_model_cont = "/data/models/crnn_H_32_W_768_E_196.pth"
device = "cpu"
num_classes = len(HWRecogIAMDataset.LABEL_2_CHAR) + 1
image_height = 32
mean_arr = np.array([[0.485, 0.456, 0.406]])
std_arr = np.array([[0.229, 0.224, 0.225]])
hw_recog_model = CRNN(num_classes, image_height)

try:
    logging.info(f"loading model from {file_model_local}")
    hw_recog_model.load_state_dict(torch.load(file_model_local, map_location=device))
except:
    logging.info(f"loading model from {file_model_cont}")
    hw_recog_model.load_state_dict(torch.load(file_model_cont, map_location=device))
hw_recog_model.to(device)
hw_recog_model.eval()


def predict_hw(img_test: np.ndarray) -> str:
    img_test = np.expand_dims(img_test, 0)
    img_test = img_test.astype(np.float32) / 255.0
    img_test = (img_test - mean_arr) / std_arr
    img_test = np.transpose(img_test, axes=[0, 3, 1, 2])
    img_tensor = torch.tensor(img_test).float()
    img_tensor = img_tensor.to(device, dtype=torch.float)
    log_probs = hw_recog_model(img_tensor)
    pred_labels = ctc_decode(log_probs.detach())
    str_pred = [HWRecogIAMDataset.LABEL_2_CHAR[i] for i in pred_labels[0]]
    str_pred = "".join(str_pred)
    return str_pred


@app.route("/predict", methods=["POST"])
def predict() -> Response:
    logging.info("IAM Handwriting recognition app")
    img_file = request.files["image_file"]
    try:
        img_arr = np.fromstring(img_file.read(), np.uint8)
    except:
        img_arr = np.fromstring(img_file.getvalue(), np.uint8)
    img_dec = cv2.imdecode(img_arr, cv2.IMREAD_COLOR)
    img_dec = cv2.cvtColor(img_dec, cv2.COLOR_BGR2RGB)

    img_dec = cv2.resize(img_dec, (768, 32), interpolation=cv2.INTER_LINEAR)

    str_pred = predict_hw(img_dec)

    dict_pred = {
        "file_name": img_file.filename,
        "prediction": str_pred,
    }
    logging.info(dict_pred)
    try:
        json_pred = jsonify(dict_pred)
    except TypeError as e:
        json_pred = jsonify({"error": str(e)})
    return json_pred


if __name__ == "__main__":
    app.run(host="0.0.0.0", debug=True, port=7860)