FORMAT_CONVERSION_MAXCOMPUTE/ocr/tools/infer/predict_det_pytorch.py

# encoding=utf8
# Copyright (c) 2020 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 io
import logging
import os
import sys
sys.path.append(os.path.dirname(os.path.abspath(__file__)) + "/../../../")
import requests
from format_convert import _global
from format_convert.utils import judge_error_code, log, namespace_to_dict, get_platform, file_lock

os.environ["FLAGS_allocator_strategy"] = 'auto_growth'
import cv2
import numpy as np
import time
import sys
os.environ['FLAGS_eager_delete_tensor_gb'] = '0'
import ocr.tools.infer.utility as utility
from ocr.ppocr.utils.logging import get_logger
from ocr.ppocr.utils.utility import get_image_file_list, check_and_read_gif
from ocr.ppocr.data import create_operators, transform
from ocr.ppocr.postprocess import build_post_process
from config.max_compute_config import MAX_COMPUTE

import torch
from torch import nn
from ocr.tools.infer.torch_det_model import DB_ResNet_18
import gc

logger = get_logger()


class TextDetector(object):
    shrink_memory_count = 0

    def __init__(self, args):
        self.args = args
        self.det_algorithm = args.det_algorithm
        pre_process_list = [{
            'DetResizeForTest': None
        }, {
            'NormalizeImage': {
                'std': [0.229, 0.224, 0.225],
                'mean': [0.485, 0.456, 0.406],
                'scale': '1./255.',
                'order': 'hwc'
            }
        }, {
            'ToCHWImage': None
        }, {
            'KeepKeys': {
                'keep_keys': ['image', 'shape']
            }
        }]
        postprocess_params = {}
        if self.det_algorithm == "DB":
            postprocess_params['name'] = 'DBPostProcess'
            postprocess_params["thresh"] = args.det_db_thresh
            postprocess_params["box_thresh"] = args.det_db_box_thresh
            postprocess_params["max_candidates"] = 1000
            postprocess_params["unclip_ratio"] = args.det_db_unclip_ratio
            postprocess_params["use_dilation"] = args.use_dilation
        elif self.det_algorithm == "EAST":
            postprocess_params['name'] = 'EASTPostProcess'
            postprocess_params["score_thresh"] = args.det_east_score_thresh
            postprocess_params["cover_thresh"] = args.det_east_cover_thresh
            postprocess_params["nms_thresh"] = args.det_east_nms_thresh
        elif self.det_algorithm == "SAST":
            pre_process_list[0] = {
                'DetResizeForTest': {
                    'resize_long': args.det_limit_side_len
                }
            }
            postprocess_params['name'] = 'SASTPostProcess'
            postprocess_params["score_thresh"] = args.det_sast_score_thresh
            postprocess_params["nms_thresh"] = args.det_sast_nms_thresh
            self.det_sast_polygon = args.det_sast_polygon
            if self.det_sast_polygon:
                postprocess_params["sample_pts_num"] = 6
                postprocess_params["expand_scale"] = 1.2
                postprocess_params["shrink_ratio_of_width"] = 0.2
            else:
                postprocess_params["sample_pts_num"] = 2
                postprocess_params["expand_scale"] = 1.0
                postprocess_params["shrink_ratio_of_width"] = 0.3
        else:
            logger.info("unknown det_algorithm:{}".format(self.det_algorithm))
            sys.exit(0)

        self.preprocess_op = create_operators(pre_process_list)
        self.postprocess_op = build_post_process(postprocess_params)

        det_model_path = args.det_model_dir
        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        model = DB_ResNet_18()
        mode_state_dict = torch.load(det_model_path, self.device)['state_dict']
        if str(self.device) == 'cpu':  # cpu处理时精度调整，加速推理
            for name, value in mode_state_dict.items():
                if get_platform() != "Windows":
                    value = value.double()
                value = torch.where((value < 1.0e-23) & (value > 0.0), 1.0e-23, value)
                value = torch.where((value > -1.0e-23) & (value < 0.0), -1.0e-23, value)
                mode_state_dict[name] = value

        model.load_state_dict(mode_state_dict)

        self.predictor = model
        self.predictor.to(self.device)
        self.predictor.eval()

        # self.predictor, self.input_tensor, self.output_tensors = utility.create_predictor(
        #     args, 'det', logger)  # paddle.jit.load(args.det_model_dir)
        # self.predictor.eval()

    def order_points_clockwise(self, pts):
        """
        reference from: https://github.com/jrosebr1/imutils/blob/master/imutils/perspective.py
        # sort the points based on their x-coordinates
        """
        xSorted = pts[np.argsort(pts[:, 0]), :]

        # grab the left-most and right-most points from the sorted
        # x-roodinate points
        leftMost = xSorted[:2, :]
        rightMost = xSorted[2:, :]

        # now, sort the left-most coordinates according to their
        # y-coordinates so we can grab the top-left and bottom-left
        # points, respectively
        leftMost = leftMost[np.argsort(leftMost[:, 1]), :]
        (tl, bl) = leftMost

        rightMost = rightMost[np.argsort(rightMost[:, 1]), :]
        (tr, br) = rightMost

        rect = np.array([tl, tr, br, bl], dtype="float32")
        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 __call__(self, img):
        # cv2.imwrite("/data2/znj/format_conversion_maxcompute/ocr/temp_image/temp.jpg",img)
        ori_im = img.copy()
        data = {'image': img}
        data = transform(data, self.preprocess_op)
        img, shape_list = data
        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()
        starttime = time.time()

        # self.input_tensor.copy_from_cpu(img)
        img = torch.from_numpy(img).float()
        img = img.to(self.device)
        try:
            # 加锁，防止太多大图片同时预测，爆显存
            if ori_im.shape[0] > 1024 and ori_im.shape[1] > 1024 and get_platform() != "Windows" and not MAX_COMPUTE:
                time2 = time.time()
                lock_file_sub = 'ocr'
                lock_file = os.path.abspath(os.path.dirname(__file__)) + "/" + lock_file_sub + ".lock"
                f = file_lock(lock_file)
                log("get file_lock " + lock_file_sub + " time " + str(time.time()-time2))
                with torch.no_grad():
                    out = self.predictor(img)
                f.close()
            else:
                with torch.no_grad():
                    out = self.predictor(img)
        except RuntimeError:
            log("ocr/tools/infer/predict_det.py predict.run error! maybe no gpu memory!")
            log("predictor shrink memory!")
            # self.predictor.clear_intermediate_tensor()
            # self.predictor.try_shrink_memory()
            if str(self.device)!='cpu':
                torch.cuda.empty_cache()
                gc.collect()
            raise RuntimeError

        # outputs = []
        # for output_tensor in self.output_tensors:
        #     output = output_tensor.copy_to_cpu()
        #     outputs.append(output)
        out = out.cpu().numpy()

        preds = {}
        preds['maps'] = out

        # if self.det_algorithm == "EAST":
        #     preds['f_geo'] = outputs[0]
        #     preds['f_score'] = outputs[1]
        # elif self.det_algorithm == 'SAST':
        #     preds['f_border'] = outputs[0]
        #     preds['f_score'] = outputs[1]
        #     preds['f_tco'] = outputs[2]
        #     preds['f_tvo'] = outputs[3]
        # elif self.det_algorithm == 'DB':
        #     preds['maps'] = outputs[0]
        # else:
        #     raise NotImplementedError
        post_result = self.postprocess_op(preds, shape_list)
        dt_boxes = post_result[0]['points']
        if self.det_algorithm == "SAST" and self.det_sast_polygon:
            dt_boxes = self.filter_tag_det_res_only_clip(dt_boxes, ori_im.shape)
        else:
            dt_boxes = self.filter_tag_det_res(dt_boxes, ori_im.shape)
        elapse = time.time() - starttime

        # 释放内存
        # print("TextDetector", self.predictor)
        # if TextDetector.shrink_memory_count % 100 == 0:
            # print("TextDetector shrink memory")
        # self.predictor.clear_intermediate_tensor()
        # self.predictor.try_shrink_memory()
        # TextDetector.shrink_memory_count += 1
        if str(self.device) != 'cpu':
            torch.cuda.empty_cache()
            # gc.collect()

        return dt_boxes, elapse


class TextDetector2(object):
    shrink_memory_count = 0

    def __init__(self, args):
        self.args = args
        self.det_algorithm = args.det_algorithm
        pre_process_list = [{
            'DetResizeForTest': None
        }, {
            'NormalizeImage': {
                'std': [0.229, 0.224, 0.225],
                'mean': [0.485, 0.456, 0.406],
                'scale': '1./255.',
                'order': 'hwc'
            }
        }, {
            'ToCHWImage': None
        }, {
            'KeepKeys': {
                'keep_keys': ['image', 'shape']
            }
        }]
        postprocess_params = {}
        if self.det_algorithm == "DB":
            postprocess_params['name'] = 'DBPostProcess'
            postprocess_params["thresh"] = args.det_db_thresh
            postprocess_params["box_thresh"] = args.det_db_box_thresh
            postprocess_params["max_candidates"] = 1000
            postprocess_params["unclip_ratio"] = args.det_db_unclip_ratio
            postprocess_params["use_dilation"] = args.use_dilation
        else:
            logger.info("unknown det_algorithm:{}".format(self.det_algorithm))
            sys.exit(0)

        self.preprocess_op = create_operators(pre_process_list)
        self.postprocess_op = build_post_process(postprocess_params)

    def order_points_clockwise(self, pts):
        """
        reference from: https://github.com/jrosebr1/imutils/blob/master/imutils/perspective.py
        # sort the points based on their x-coordinates
        """
        xSorted = pts[np.argsort(pts[:, 0]), :]

        # grab the left-most and right-most points from the sorted
        # x-roodinate points
        leftMost = xSorted[:2, :]
        rightMost = xSorted[2:, :]

        # now, sort the left-most coordinates according to their
        # y-coordinates so we can grab the top-left and bottom-left
        # points, respectively
        leftMost = leftMost[np.argsort(leftMost[:, 1]), :]
        (tl, bl) = leftMost

        rightMost = rightMost[np.argsort(rightMost[:, 1]), :]
        (tr, br) = rightMost

        rect = np.array([tl, tr, br, bl], dtype="float32")
        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 __call__(self, img):
        from format_convert.convert_need_interface import from_gpu_interface_redis
        # 预处理
        ori_im = img.copy()
        data = {'image': img}
        data = transform(data, self.preprocess_op)
        img, shape_list = data
        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()
        starttime = time.time()

        # # 压缩numpy
        # compressed_array = io.BytesIO()
        # np.savez_compressed(compressed_array, img)
        # compressed_array.seek(0)
        # img = compressed_array.read()

        # 调用GPU接口
        _dict = {"inputs": img, "args": str(namespace_to_dict(self.args)), "md5": _global.get("md5")}
        result = from_gpu_interface_redis(_dict, model_type="ocr", predictor_type="det")
        if judge_error_code(result):
            logging.error("from_gpu_interface failed! " + str(result))
            raise requests.exceptions.RequestException

        _preds = result.get("preds")
        gpu_time = result.get("gpu_time")

        # # 解压numpy
        # decompressed_array = io.BytesIO()
        # decompressed_array.write(_preds)
        # decompressed_array.seek(0)
        # _preds = np.load(decompressed_array, allow_pickle=True)['arr_0']
        # log("inputs.shape" + str(_preds.shape))

        # 后处理
        preds = {}
        if self.det_algorithm == 'DB':
            preds['maps'] = _preds
        else:
            raise NotImplementedError

        post_result = self.postprocess_op(preds, shape_list)
        dt_boxes = post_result[0]['points']
        dt_boxes = self.filter_tag_det_res(dt_boxes, ori_im.shape)
        elapse = time.time() - starttime

        log("ocr model predict time - det - time " + str(gpu_time))
        return dt_boxes, elapse


if __name__ == "__main__":
    args = utility.parse_args()
    image_file_list = get_image_file_list(args.image_dir)
    text_detector = TextDetector(args)
    count = 0
    total_time = 0
    draw_img_save = "./inference_results"
    if not os.path.exists(draw_img_save):
        os.makedirs(draw_img_save)
    for image_file in image_file_list:
        img, flag = check_and_read_gif(image_file)
        if not flag:
            img = cv2.imread(image_file)
        if img is None:
            logger.info("error in loading image:{}".format(image_file))
            continue
        dt_boxes, elapse = text_detector(img)
        if count > 0:
            total_time += elapse
        count += 1
        logger.info("Predict time of {}: {}".format(image_file, elapse))
        src_im = utility.draw_text_det_res(dt_boxes, image_file)
        img_name_pure = os.path.split(image_file)[-1]
        img_path = os.path.join(draw_img_save,
                                "det_res_{}".format(img_name_pure))
        cv2.imwrite(img_path, src_im)
        logger.info("The visualized image saved in {}".format(img_path))
    if count > 1:
        logger.info("Avg Time: {}".format(total_time / (count - 1)))