vechile_plate.py

# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import os
import yaml
import glob
from functools import reduce

import time
import cv2
import numpy as np
import math
import paddle

import sys
# add deploy path of PadleDetection to sys.path
parent_path = os.path.abspath(os.path.join(__file__, *(['..'])))
sys.path.insert(0, parent_path)

from utils import Timer, get_current_memory_mb
from infer import Detector, get_test_images, print_arguments, create_inputs
from vechile_plateutils import create_predictor, get_infer_gpuid, argsparser, get_rotate_crop_image
from vecplatepostprocess import build_post_process
from preprocess import preprocess, Resize, NormalizeImage, Permute, PadStride, LetterBoxResize, WarpAffine, Pad, decode_image, Resize_Mult32


class PlateDetector(object):
    def __init__(self, args):
        self.args = args
        self.det_algorithm = args.det_algorithm
        self.pre_process_list = {
            'Resize_Mult32': {
                'limit_side_len': args.det_limit_side_len,
                'limit_type': args.det_limit_type,
            },
            'NormalizeImage': {
                'mean': [0.485, 0.456, 0.406],
                'std': [0.229, 0.224, 0.225],
                'is_scale': True,
            },
            'Permute': {}
        }
        postprocess_params = {}
        postprocess_params['name'] = 'DBPostProcess'
        postprocess_params["thresh"] = 0.3
        postprocess_params["box_thresh"] = 0.6
        postprocess_params["max_candidates"] = 1000
        postprocess_params["unclip_ratio"] = 1.5
        postprocess_params["use_dilation"] = False
        postprocess_params["score_mode"] = "fast"

        self.postprocess_op = build_post_process(postprocess_params)
        self.predictor, self.input_tensor, self.output_tensors, self.config = create_predictor(
            args, 'det')

        if args.run_benchmark:
            import auto_log
            pid = os.getpid()
            gpu_id = get_infer_gpuid()
            self.autolog = auto_log.AutoLogger(
                model_name="det",
                model_precision="fp32",
                batch_size=1,
                data_shape="dynamic",
                save_path=None,
                inference_config=self.config,
                pids=pid,
                process_name=None,
                gpu_ids=gpu_id if args.device == "GPU" else None,
                time_keys=[
                    'preprocess_time', 'inference_time', 'postprocess_time'
                ],
                warmup=2, )

    def preprocess(self, image_list):
        preprocess_ops = []
        for op_type, new_op_info in self.pre_process_list.items():
            preprocess_ops.append(eval(op_type)(**new_op_info))

        input_im_lst = []
        input_im_info_lst = []
        for im_path in image_list:
            im, im_info = preprocess(im_path, preprocess_ops)
            input_im_lst.append(im)
            input_im_info_lst.append(im_info['im_shape'])

        return np.stack(input_im_lst, axis=0), input_im_info_lst

    def order_points_clockwise(self, pts):
        rect = np.zeros((4, 2), dtype="float32")
        s = pts.sum(axis=1)
        rect[0] = pts[np.argmin(s)]
        rect[2] = pts[np.argmax(s)]
        diff = np.diff(pts, axis=1)
        rect[1] = pts[np.argmin(diff)]
        rect[3] = pts[np.argmax(diff)]
        return rect

    def clip_det_res(self, points, img_height, img_width):
        for pno in range(points.shape[0]):
            points[pno, 0] = int(min(max(points[pno, 0], 0), img_width - 1))
            points[pno, 1] = int(min(max(points[pno, 1], 0), img_height - 1))
        return points

    def filter_tag_det_res(self, dt_boxes, image_shape):
        img_height, img_width = image_shape[0:2]
        dt_boxes_new = []
        for box in dt_boxes:
            box = self.order_points_clockwise(box)
            box = self.clip_det_res(box, img_height, img_width)
            rect_width = int(np.linalg.norm(box[0] - box[1]))
            rect_height = int(np.linalg.norm(box[0] - box[3]))
            if rect_width <= 3 or rect_height <= 3:
                continue
            dt_boxes_new.append(box)
        dt_boxes = np.array(dt_boxes_new)
        return dt_boxes

    def filter_tag_det_res_only_clip(self, dt_boxes, image_shape):
        img_height, img_width = image_shape[0:2]
        dt_boxes_new = []
        for box in dt_boxes:
            box = self.clip_det_res(box, img_height, img_width)
            dt_boxes_new.append(box)
        dt_boxes = np.array(dt_boxes_new)
        return dt_boxes

    def predict_image(self, img):
        st = time.time()

        if self.args.run_benchmark:
            self.autolog.times.start()

        img, shape_list = self.preprocess(img)
        if img is None:
            return None, 0
        # img = np.expand_dims(img, axis=0)
        # shape_list = np.expand_dims(shape_list, axis=0)
        # img = img.copy()

        if self.args.run_benchmark:
            self.autolog.times.stamp()

        self.input_tensor.copy_from_cpu(img)
        self.predictor.run()
        outputs = []
        for output_tensor in self.output_tensors:
            output = output_tensor.copy_to_cpu()
            outputs.append(output)
        if self.args.run_benchmark:
            self.autolog.times.stamp()

        preds = {}
        preds['maps'] = outputs[0]

        #self.predictor.try_shrink_memory()
        post_result = self.postprocess_op(preds, shape_list)
        dt_boxes = post_result[0]['points']
        dt_boxes = self.filter_tag_det_res(dt_boxes, shape_list[0])

        if self.args.run_benchmark:
            self.autolog.times.end(stamp=True)
        et = time.time()
        return dt_boxes, et - st


class TextRecognizer(object):
    def __init__(self,
                 FLAGS,
                 input_shape=[3, 48, 320],
                 batch_size=8,
                 rec_algorithm="SVTR",
                 word_dict_path="rec_word_dict.txt",
                 use_gpu=True,
                 benchmark=False):
        self.rec_image_shape = input_shape
        self.rec_batch_num = batch_size
        self.rec_algorithm = rec_algorithm
        isuse_space_char = True

        postprocess_params = {
            'name': 'CTCLabelDecode',
            "character_dict_path": word_dict_path,
            "use_space_char": isuse_space_char
        }
        if self.rec_algorithm == "SRN":
            postprocess_params = {
                'name': 'SRNLabelDecode',
                "character_dict_path": word_dict_path,
                "use_space_char": isuse_space_char
            }
        elif self.rec_algorithm == "RARE":
            postprocess_params = {
                'name': 'AttnLabelDecode',
                "character_dict_path": word_dict_path,
                "use_space_char": isuse_space_char
            }
        elif self.rec_algorithm == 'NRTR':
            postprocess_params = {
                'name': 'NRTRLabelDecode',
                "character_dict_path": word_dict_path,
                "use_space_char": isuse_space_char
            }
        elif self.rec_algorithm == "SAR":
            postprocess_params = {
                'name': 'SARLabelDecode',
                "character_dict_path": word_dict_path,
                "use_space_char": isuse_space_char
            }
        self.postprocess_op = build_post_process(postprocess_params)
        self.predictor, self.input_tensor, self.output_tensors, self.config = \
            create_predictor(FLAGS, 'rec')
        self.benchmark = benchmark
        self.use_onnx = False
        if benchmark:
            import auto_log
            pid = os.getpid()
            gpu_id = get_infer_gpuid()
            self.autolog = auto_log.AutoLogger(
                model_name="rec",
                model_precision='fp32',
                batch_size=batch_size,
                data_shape="dynamic",
                save_path=None,  #save_log_path,
                inference_config=self.config,
                pids=pid,
                process_name=None,
                gpu_ids=gpu_id if use_gpu else None,
                time_keys=[
                    'preprocess_time', 'inference_time', 'postprocess_time'
                ],
                warmup=0)

    def resize_norm_img(self, img, max_wh_ratio):
        imgC, imgH, imgW = self.rec_image_shape
        if self.rec_algorithm == 'NRTR':
            img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
            # return padding_im
            image_pil = Image.fromarray(np.uint8(img))
            img = image_pil.resize([100, 32], Image.ANTIALIAS)
            img = np.array(img)
            norm_img = np.expand_dims(img, -1)
            norm_img = norm_img.transpose((2, 0, 1))
            return norm_img.astype(np.float32) / 128. - 1.

        assert imgC == img.shape[2]
        imgW = int((imgH * max_wh_ratio))
        if self.use_onnx:
            w = self.input_tensor.shape[3:][0]
            if w is not None and w > 0:
                imgW = w

        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))
        if self.rec_algorithm == 'RARE':
            if resized_w > self.rec_image_shape[2]:
                resized_w = self.rec_image_shape[2]
            imgW = self.rec_image_shape[2]
        resized_image = cv2.resize(img, (resized_w, imgH))
        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 resize_norm_img_svtr(self, img, image_shape):

        imgC, imgH, imgW = image_shape
        resized_image = cv2.resize(
            img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
        resized_image = resized_image.astype('float32')
        resized_image = resized_image.transpose((2, 0, 1)) / 255
        resized_image -= 0.5
        resized_image /= 0.5
        return resized_image

    def resize_norm_img_srn(self, img, image_shape):
        imgC, imgH, imgW = image_shape

        img_black = np.zeros((imgH, imgW))
        im_hei = img.shape[0]
        im_wid = img.shape[1]

        if im_wid <= im_hei * 1:
            img_new = cv2.resize(img, (imgH * 1, imgH))
        elif im_wid <= im_hei * 2:
            img_new = cv2.resize(img, (imgH * 2, imgH))
        elif im_wid <= im_hei * 3:
            img_new = cv2.resize(img, (imgH * 3, imgH))
        else:
            img_new = cv2.resize(img, (imgW, imgH))

        img_np = np.asarray(img_new)
        img_np = cv2.cvtColor(img_np, cv2.COLOR_BGR2GRAY)
        img_black[:, 0:img_np.shape[1]] = img_np
        img_black = img_black[:, :, np.newaxis]

        row, col, c = img_black.shape
        c = 1

        return np.reshape(img_black, (c, row, col)).astype(np.float32)

    def srn_other_inputs(self, image_shape, num_heads, max_text_length):

        imgC, imgH, imgW = image_shape
        feature_dim = int((imgH / 8) * (imgW / 8))

        encoder_word_pos = np.array(range(0, feature_dim)).reshape(
            (feature_dim, 1)).astype('int64')
        gsrm_word_pos = np.array(range(0, max_text_length)).reshape(
            (max_text_length, 1)).astype('int64')

        gsrm_attn_bias_data = np.ones((1, max_text_length, max_text_length))
        gsrm_slf_attn_bias1 = np.triu(gsrm_attn_bias_data, 1).reshape(
            [-1, 1, max_text_length, max_text_length])
        gsrm_slf_attn_bias1 = np.tile(
            gsrm_slf_attn_bias1,
            [1, num_heads, 1, 1]).astype('float32') * [-1e9]

        gsrm_slf_attn_bias2 = np.tril(gsrm_attn_bias_data, -1).reshape(
            [-1, 1, max_text_length, max_text_length])
        gsrm_slf_attn_bias2 = np.tile(
            gsrm_slf_attn_bias2,
            [1, num_heads, 1, 1]).astype('float32') * [-1e9]

        encoder_word_pos = encoder_word_pos[np.newaxis, :]
        gsrm_word_pos = gsrm_word_pos[np.newaxis, :]

        return [
            encoder_word_pos, gsrm_word_pos, gsrm_slf_attn_bias1,
            gsrm_slf_attn_bias2
        ]

    def process_image_srn(self, img, image_shape, num_heads, max_text_length):
        norm_img = self.resize_norm_img_srn(img, image_shape)
        norm_img = norm_img[np.newaxis, :]

        [encoder_word_pos, gsrm_word_pos, gsrm_slf_attn_bias1, gsrm_slf_attn_bias2] = \
            self.srn_other_inputs(image_shape, num_heads, max_text_length)

        gsrm_slf_attn_bias1 = gsrm_slf_attn_bias1.astype(np.float32)
        gsrm_slf_attn_bias2 = gsrm_slf_attn_bias2.astype(np.float32)
        encoder_word_pos = encoder_word_pos.astype(np.int64)
        gsrm_word_pos = gsrm_word_pos.astype(np.int64)

        return (norm_img, encoder_word_pos, gsrm_word_pos, gsrm_slf_attn_bias1,
                gsrm_slf_attn_bias2)

    def resize_norm_img_sar(self, img, image_shape,
                            width_downsample_ratio=0.25):
        imgC, imgH, imgW_min, imgW_max = image_shape
        h = img.shape[0]
        w = img.shape[1]
        valid_ratio = 1.0
        # make sure new_width is an integral multiple of width_divisor.
        width_divisor = int(1 / width_downsample_ratio)
        # resize
        ratio = w / float(h)
        resize_w = math.ceil(imgH * ratio)
        if resize_w % width_divisor != 0:
            resize_w = round(resize_w / width_divisor) * width_divisor
        if imgW_min is not None:
            resize_w = max(imgW_min, resize_w)
        if imgW_max is not None:
            valid_ratio = min(1.0, 1.0 * resize_w / imgW_max)
            resize_w = min(imgW_max, resize_w)
        resized_image = cv2.resize(img, (resize_w, imgH))
        resized_image = resized_image.astype('float32')
        # norm 
        if image_shape[0] == 1:
            resized_image = resized_image / 255
            resized_image = resized_image[np.newaxis, :]
        else:
            resized_image = resized_image.transpose((2, 0, 1)) / 255
        resized_image -= 0.5
        resized_image /= 0.5
        resize_shape = resized_image.shape
        padding_im = -1.0 * np.ones((imgC, imgH, imgW_max), dtype=np.float32)
        padding_im[:, :, 0:resize_w] = resized_image
        pad_shape = padding_im.shape

        return padding_im, resize_shape, pad_shape, valid_ratio

    def __call__(self, img_list):
        img_num = len(img_list)
        # Calculate the aspect ratio of all text bars
        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
        st = time.time()
        if self.benchmark:
            self.autolog.times.start()
        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
            max_wh_ratio = imgW / imgH
            # max_wh_ratio = 0
            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):

                if self.rec_algorithm == "SAR":
                    norm_img, _, _, valid_ratio = self.resize_norm_img_sar(
                        img_list[indices[ino]], self.rec_image_shape)
                    norm_img = norm_img[np.newaxis, :]
                    valid_ratio = np.expand_dims(valid_ratio, axis=0)
                    valid_ratios = []
                    valid_ratios.append(valid_ratio)
                    norm_img_batch.append(norm_img)
                elif self.rec_algorithm == "SRN":
                    norm_img = self.process_image_srn(
                        img_list[indices[ino]], self.rec_image_shape, 8, 25)
                    encoder_word_pos_list = []
                    gsrm_word_pos_list = []
                    gsrm_slf_attn_bias1_list = []
                    gsrm_slf_attn_bias2_list = []
                    encoder_word_pos_list.append(norm_img[1])
                    gsrm_word_pos_list.append(norm_img[2])
                    gsrm_slf_attn_bias1_list.append(norm_img[3])
                    gsrm_slf_attn_bias2_list.append(norm_img[4])
                    norm_img_batch.append(norm_img[0])
                elif self.rec_algorithm == "SVTR":
                    norm_img = self.resize_norm_img_svtr(img_list[indices[ino]],
                                                         self.rec_image_shape)
                    norm_img = norm_img[np.newaxis, :]
                    norm_img_batch.append(norm_img)
                else:
                    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()
            if self.benchmark:
                self.autolog.times.stamp()

            if self.rec_algorithm == "SRN":
                encoder_word_pos_list = np.concatenate(encoder_word_pos_list)
                gsrm_word_pos_list = np.concatenate(gsrm_word_pos_list)
                gsrm_slf_attn_bias1_list = np.concatenate(
                    gsrm_slf_attn_bias1_list)
                gsrm_slf_attn_bias2_list = np.concatenate(
                    gsrm_slf_attn_bias2_list)

                inputs = [
                    norm_img_batch,
                    encoder_word_pos_list,
                    gsrm_word_pos_list,
                    gsrm_slf_attn_bias1_list,
                    gsrm_slf_attn_bias2_list,
                ]
                if self.use_onnx:
                    input_dict = {}
                    input_dict[self.input_tensor.name] = norm_img_batch
                    outputs = self.predictor.run(self.output_tensors,
                                                 input_dict)
                    preds = {"predict": outputs[2]}
                else:
                    input_names = self.predictor.get_input_names()
                    for i in range(len(input_names)):
                        input_tensor = self.predictor.get_input_handle(
                            input_names[i])
                        input_tensor.copy_from_cpu(inputs[i])
                    self.predictor.run()
                    outputs = []
                    for output_tensor in self.output_tensors:
                        output = output_tensor.copy_to_cpu()
                        outputs.append(output)
                    if self.benchmark:
                        self.autolog.times.stamp()
                    preds = {"predict": outputs[2]}
            elif self.rec_algorithm == "SAR":
                valid_ratios = np.concatenate(valid_ratios)
                inputs = [
                    norm_img_batch,
                    valid_ratios,
                ]
                if self.use_onnx:
                    input_dict = {}
                    input_dict[self.input_tensor.name] = norm_img_batch
                    outputs = self.predictor.run(self.output_tensors,
                                                 input_dict)
                    preds = outputs[0]
                else:
                    input_names = self.predictor.get_input_names()
                    for i in range(len(input_names)):
                        input_tensor = self.predictor.get_input_handle(
                            input_names[i])
                        input_tensor.copy_from_cpu(inputs[i])
                    self.predictor.run()
                    outputs = []
                    for output_tensor in self.output_tensors:
                        output = output_tensor.copy_to_cpu()
                        outputs.append(output)
                    if self.benchmark:
                        self.autolog.times.stamp()
                    preds = outputs[0]
            else:
                if self.use_onnx:
                    input_dict = {}
                    input_dict[self.input_tensor.name] = norm_img_batch
                    outputs = self.predictor.run(self.output_tensors,
                                                 input_dict)
                    preds = outputs[0]
                else:
                    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 self.benchmark:
                        self.autolog.times.stamp()
                    if len(outputs) != 1:
                        preds = outputs
                    else:
                        preds = outputs[0]
            rec_result = self.postprocess_op(preds)
            for rno in range(len(rec_result)):
                rec_res[indices[beg_img_no + rno]] = rec_result[rno]
            if self.benchmark:
                self.autolog.times.end(stamp=True)
        return rec_res, time.time() - st


class PlateRecognizer(object):
    def __init__(self):
        self.batch_size = 8
        self.platedetector = PlateDetector(FLAGS)
        self.textrecognizer = TextRecognizer(
            FLAGS,
            input_shape=[3, 48, 320],
            batch_size=8,
            rec_algorithm="SVTR",
            word_dict_path="rec_word_dict.txt",
            use_gpu=True,
            benchmark=False)

    def get_platelicense(self, image_list):
        plate_text_list = []
        plateboxes, det_time = self.platedetector.predict_image(image_list)
        for idx, boxes_pcar in enumerate(plateboxes):
            plate_images = get_rotate_crop_image(image_list[idx], boxes_pcar)
            print(plate_images.shape)
            plate_texts = self.textrecognizer(plate_images)
            plate_text_list.append(plate_texts)
            import pdb
            pdb.set_trace()
        return results


def main():
    detector = PlateRecognizer()
    # predict from image
    if FLAGS.image_dir is None and FLAGS.image_file is not None:
        assert FLAGS.batch_size == 1, "batch_size should be 1, when image_file is not None"
    img_list = get_test_images(FLAGS.image_dir, FLAGS.image_file)
    for img in img_list:
        image = cv2.imread(img)
        results = detector.get_platelicense([image])
    if not FLAGS.run_benchmark:
        detector.det_times.info(average=True)
    else:
        mems = {
            'cpu_rss_mb': detector.cpu_mem / len(img_list),
            'gpu_rss_mb': detector.gpu_mem / len(img_list),
            'gpu_util': detector.gpu_util * 100 / len(img_list)
        }

        perf_info = detector.det_times.report(average=True)
        model_dir = FLAGS.model_dir
        mode = FLAGS.run_mode
        model_info = {
            'model_name': model_dir.strip('/').split('/')[-1],
            'precision': mode.split('_')[-1]
        }
        data_info = {
            'batch_size': FLAGS.batch_size,
            'shape': "dynamic_shape",
            'data_num': perf_info['img_num']
        }
        det_log = PaddleInferBenchmark(detector.config, model_info, data_info,
                                       perf_info, mems)
        det_log('Attr')


if __name__ == '__main__':
    paddle.enable_static()
    parser = argsparser()
    FLAGS = parser.parse_args()
    print_arguments(FLAGS)
    FLAGS.device = FLAGS.device.upper()
    assert FLAGS.device in ['CPU', 'GPU', 'XPU'
                            ], "device should be CPU, GPU or XPU"
    assert not FLAGS.use_gpu, "use_gpu has been deprecated, please use --device"

    main()