All/Python/PaddleOCR/hmOCR/parts/recognizer.py

import cv2
import math
import numpy as np
from .utils import *
from hmOCR.argument import Args
from .operator import CTCLabelDecode


class Recognizer:
    def __init__(self, args: "Args"):
        self.rec_image_shape = [int(v) for v in args.rec_image_shape.split(",")]
        self.rec_batch_num = args.rec_batch_num
        # rec_algorithm: only "SVTR_LCNet" now
        # self.rec_algorithm = args.rec_algorithm

        self.post_op = CTCLabelDecode(args.rec_char_dict_path)
        self.predictor, self.input_tensor, self.output_tensors = create_predictor(args, "rec")

    def resize_norm_img(self, img, max_wh_ratio):
        imgC, imgH, imgW = self.rec_image_shape
        assert imgC == img.shape[2]
        imgW = int((imgH * max_wh_ratio))

        h, w = img.shape[:2]
        ratio = w / float(h)
        if math.ceil(imgH * ratio) > imgW:
            resized_w = imgW
        else:
            resized_w = int(math.ceil(imgH * ratio))

        resized_image = cv2.resize(img, (resized_w, imgH))  # noqa
        resized_image = resized_image.astype("float32")
        resized_image = resized_image.transpose((2, 0, 1)) / 255
        resized_image -= 0.5
        resized_image /= 0.5
        padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
        padding_im[:, :, 0:resized_w] = resized_image
        return padding_im

    def __call__(self, img_list, use_space=False):
        img_num = len(img_list)
        width_list = []
        for img in img_list:
            width_list.append(img.shape[1] / float(img.shape[0]))
        # Sorting can speed up the recognition process
        indices = np.argsort(np.array(width_list))
        rec_res = [["", 0.0]] * img_num
        batch_num = self.rec_batch_num

        for beg_img_no in range(0, img_num, batch_num):
            end_img_no = min(img_num, beg_img_no + batch_num)
            norm_img_batch = []

            imgC, imgH, imgW = self.rec_image_shape[:3]
            max_wh_ratio = imgW / imgH
            for ino in range(beg_img_no, end_img_no):
                h, w = img_list[indices[ino]].shape[0:2]
                wh_ratio = w * 1.0 / h
                max_wh_ratio = max(max_wh_ratio, wh_ratio)
            for ino in range(beg_img_no, end_img_no):
                norm_img = self.resize_norm_img(img_list[indices[ino]], max_wh_ratio)
                norm_img = norm_img[np.newaxis, :]
                norm_img_batch.append(norm_img)
            norm_img_batch = np.concatenate(norm_img_batch)
            norm_img_batch = norm_img_batch.copy()

            self.input_tensor.copy_from_cpu(norm_img_batch)
            self.predictor.run()
            outputs = []
            for output_tensor in self.output_tensors:
                output = output_tensor.copy_to_cpu()
                outputs.append(output)
            if len(outputs) != 1:
                preds = outputs
            else:
                preds = outputs[0]
            rec_result = self.post_op(preds, use_space=use_space)
            for rno in range(len(rec_result)):
                rec_res[indices[beg_img_no + rno]] = rec_result[rno]

        return rec_res