Huawei_Technology
/
mindspore

# Copyright 2020 Huawei Technologies Co., Ltd
#
# 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.
# ============================================================================
"""YoloV4 eval."""
import os
import argparse
import datetime
import time
import sys
from collections import defaultdict

import numpy as np
from pycocotools.coco import COCO
from pycocotools.cocoeval import COCOeval

from mindspore import Tensor
from mindspore.context import ParallelMode
from mindspore import context
from mindspore.train.serialization import load_checkpoint, load_param_into_net
import mindspore as ms

from src.yolo import YOLOV4CspDarkNet53
from src.logger import get_logger
from src.yolo_dataset import create_yolo_dataset
from src.config import ConfigYOLOV4CspDarkNet53

parser = argparse.ArgumentParser('mindspore coco testing')

# device related
parser.add_argument('--device_target', type=str, default='Ascend',
                    help='device where the code will be implemented. (Default: Ascend)')

# dataset related
parser.add_argument('--data_dir', type=str, default='', help='train data dir')
parser.add_argument('--per_batch_size', default=1, type=int, help='batch size for per gpu')

# network related
parser.add_argument('--pretrained', default='', type=str, help='model_path, local pretrained model to load')

# logging related
parser.add_argument('--log_path', type=str, default='outputs/', help='checkpoint save location')

# detect_related
parser.add_argument('--nms_thresh', type=float, default=0.5, help='threshold for NMS')
parser.add_argument('--ann_file', type=str, default='', help='path to annotation')
parser.add_argument('--testing_shape', type=str, default='', help='shape for test ')
parser.add_argument('--ignore_threshold', type=float, default=0.001, help='threshold to throw low quality boxes')

args, _ = parser.parse_known_args()

args.data_root = os.path.join(args.data_dir, 'val2017')
args.ann_file = os.path.join(args.data_dir, 'annotations/instances_val2017.json')

class Redirct:
    def __init__(self):
        self.content = ""

    def write(self, content):
        self.content += content

    def flush(self):
        self.content = ""


class DetectionEngine:
    """Detection engine."""
    def __init__(self, args_detection):
        self.ignore_threshold = args_detection.ignore_threshold
        self.labels = ['person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus', 'train', 'truck', 'boat',
                       'traffic light', 'fire hydrant', 'stop sign', 'parking meter', 'bench', 'bird', 'cat',
                       'dog', 'horse', 'sheep', 'cow', 'elephant', 'bear', 'zebra', 'giraffe', 'backpack',
                       'umbrella', 'handbag', 'tie', 'suitcase', 'frisbee', 'skis', 'snowboard', 'sports ball',
                       'kite', 'baseball bat', 'baseball glove', 'skateboard', 'surfboard', 'tennis racket',
                       'bottle', 'wine glass', 'cup', 'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple',
                       'sandwich', 'orange', 'broccoli', 'carrot', 'hot dog', 'pizza', 'donut', 'cake', 'chair',
                       'couch', 'potted plant', 'bed', 'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote',
                       'keyboard', 'cell phone', 'microwave', 'oven', 'toaster', 'sink', 'refrigerator', 'book',
                       'clock', 'vase', 'scissors', 'teddy bear', 'hair drier', 'toothbrush']
        self.num_classes = len(self.labels)
        self.results = {}
        self.file_path = ''
        self.save_prefix = args_detection.outputs_dir
        self.ann_file = args_detection.ann_file
        self._coco = COCO(self.ann_file)
        self._img_ids = list(sorted(self._coco.imgs.keys()))
        self.det_boxes = []
        self.nms_thresh = args_detection.nms_thresh
        self.coco_catids = self._coco.getCatIds()

    def do_nms_for_results(self):
        """Get result boxes."""
        for img_id in self.results:
            for clsi in self.results[img_id]:
                dets = self.results[img_id][clsi]
                dets = np.array(dets)
                keep_index = self._diou_nms(dets, thresh=0.6)

                keep_box = [{'image_id': int(img_id),
                             'category_id': int(clsi),
                             'bbox': list(dets[i][:4].astype(float)),
                             'score': dets[i][4].astype(float)}
                            for i in keep_index]
                self.det_boxes.extend(keep_box)

    def _nms(self, predicts, threshold):
        """Calculate NMS."""
        # convert xywh -> xmin ymin xmax ymax
        x1 = predicts[:, 0]
        y1 = predicts[:, 1]
        x2 = x1 + predicts[:, 2]
        y2 = y1 + predicts[:, 3]
        scores = predicts[:, 4]

        areas = (x2 - x1 + 1) * (y2 - y1 + 1)
        order = scores.argsort()[::-1]

        reserved_boxes = []
        while order.size > 0:
            i = order[0]
            reserved_boxes.append(i)
            max_x1 = np.maximum(x1[i], x1[order[1:]])
            max_y1 = np.maximum(y1[i], y1[order[1:]])
            min_x2 = np.minimum(x2[i], x2[order[1:]])
            min_y2 = np.minimum(y2[i], y2[order[1:]])

            intersect_w = np.maximum(0.0, min_x2 - max_x1 + 1)
            intersect_h = np.maximum(0.0, min_y2 - max_y1 + 1)
            intersect_area = intersect_w * intersect_h
            ovr = intersect_area / (areas[i] + areas[order[1:]] - intersect_area)

            indexes = np.where(ovr <= threshold)[0]
            order = order[indexes + 1]
        return reserved_boxes

    def _diou_nms(self, dets, thresh=0.5):
        """
        convert xywh -> xmin ymin xmax ymax
        """
        x1 = dets[:, 0]
        y1 = dets[:, 1]
        x2 = x1 + dets[:, 2]
        y2 = y1 + dets[:, 3]
        scores = dets[:, 4]
        areas = (x2 - x1 + 1) * (y2 - y1 + 1)
        order = scores.argsort()[::-1]
        keep = []
        while order.size > 0:
            i = order[0]
            keep.append(i)
            xx1 = np.maximum(x1[i], x1[order[1:]])
            yy1 = np.maximum(y1[i], y1[order[1:]])
            xx2 = np.minimum(x2[i], x2[order[1:]])
            yy2 = np.minimum(y2[i], y2[order[1:]])

            w = np.maximum(0.0, xx2 - xx1 + 1)
            h = np.maximum(0.0, yy2 - yy1 + 1)
            inter = w * h
            ovr = inter / (areas[i] + areas[order[1:]] - inter)
            center_x1 = (x1[i] + x2[i]) / 2
            center_x2 = (x1[order[1:]] + x2[order[1:]]) / 2
            center_y1 = (y1[i] + y2[i]) / 2
            center_y2 = (y1[order[1:]] + y2[order[1:]]) / 2
            inter_diag = (center_x2 - center_x1) ** 2 + (center_y2 - center_y1) ** 2
            out_max_x = np.maximum(x2[i], x2[order[1:]])
            out_max_y = np.maximum(y2[i], y2[order[1:]])
            out_min_x = np.minimum(x1[i], x1[order[1:]])
            out_min_y = np.minimum(y1[i], y1[order[1:]])
            outer_diag = (out_max_x - out_min_x) ** 2 + (out_max_y - out_min_y) ** 2
            diou = ovr - inter_diag / outer_diag
            diou = np.clip(diou, -1, 1)
            inds = np.where(diou <= thresh)[0]
            order = order[inds + 1]
        return keep


    def write_result(self):
        """Save result to file."""
        import json
        t = datetime.datetime.now().strftime('_%Y_%m_%d_%H_%M_%S')
        try:
            self.file_path = self.save_prefix + '/predict' + t + '.json'
            f = open(self.file_path, 'w')
            json.dump(self.det_boxes, f)
        except IOError as e:
            raise RuntimeError("Unable to open json file to dump. What(): {}".format(str(e)))
        else:
            f.close()
            return self.file_path

    def get_eval_result(self):
        """Get eval result."""
        coco_gt = COCO(self.ann_file)
        coco_dt = coco_gt.loadRes(self.file_path)
        coco_eval = COCOeval(coco_gt, coco_dt, 'bbox')
        coco_eval.evaluate()
        coco_eval.accumulate()
        rdct = Redirct()
        stdout = sys.stdout
        sys.stdout = rdct
        coco_eval.summarize()
        sys.stdout = stdout
        return rdct.content

    def detect(self, outputs, batch, image_shape, image_id):
        """Detect boxes."""
        outputs_num = len(outputs)
        # output [|32, 52, 52, 3, 85| ]
        for batch_id in range(batch):
            for out_id in range(outputs_num):
                # 32, 52, 52, 3, 85
                out_item = outputs[out_id]
                # 52, 52, 3, 85
                out_item_single = out_item[batch_id, :]
                # get number of items in one head, [B, gx, gy, anchors, 5+80]
                dimensions = out_item_single.shape[:-1]
                out_num = 1
                for d in dimensions:
                    out_num *= d
                ori_w, ori_h = image_shape[batch_id]
                img_id = int(image_id[batch_id])
                x = out_item_single[..., 0] * ori_w
                y = out_item_single[..., 1] * ori_h
                w = out_item_single[..., 2] * ori_w
                h = out_item_single[..., 3] * ori_h

                conf = out_item_single[..., 4:5]
                cls_emb = out_item_single[..., 5:]

                cls_argmax = np.expand_dims(np.argmax(cls_emb, axis=-1), axis=-1)
                x = x.reshape(-1)
                y = y.reshape(-1)
                w = w.reshape(-1)
                h = h.reshape(-1)
                cls_emb = cls_emb.reshape(-1, self.num_classes)
                conf = conf.reshape(-1)
                cls_argmax = cls_argmax.reshape(-1)

                x_top_left = x - w / 2.
                y_top_left = y - h / 2.
                # create all False
                flag = np.random.random(cls_emb.shape) > sys.maxsize
                for i in range(flag.shape[0]):
                    c = cls_argmax[i]
                    flag[i, c] = True
                confidence = cls_emb[flag] * conf
                for x_lefti, y_lefti, wi, hi, confi, clsi in zip(x_top_left, y_top_left, w, h, confidence, cls_argmax):
                    if confi < self.ignore_threshold:
                        continue
                    if img_id not in self.results:
                        self.results[img_id] = defaultdict(list)
                    x_lefti = max(0, x_lefti)
                    y_lefti = max(0, y_lefti)
                    wi = min(wi, ori_w)
                    hi = min(hi, ori_h)
                    # transform catId to match coco
                    coco_clsi = self.coco_catids[clsi]
                    self.results[img_id][coco_clsi].append([x_lefti, y_lefti, wi, hi, confi])


def convert_testing_shape(args_testing_shape):
    """Convert testing shape to list."""
    testing_shape = [int(args_testing_shape), int(args_testing_shape)]
    return testing_shape


if __name__ == "__main__":
    start_time = time.time()
    device_id = int(os.getenv('DEVICE_ID')) if os.getenv('DEVICE_ID') else 0
    context.set_context(mode=context.GRAPH_MODE, device_target=args.device_target, device_id=device_id)

    # logger
    args.outputs_dir = os.path.join(args.log_path,
                                    datetime.datetime.now().strftime('%Y-%m-%d_time_%H_%M_%S'))
    rank_id = int(os.environ.get('RANK_ID')) if os.environ.get('RANK_ID') else 0
    args.logger = get_logger(args.outputs_dir, rank_id)

    context.reset_auto_parallel_context()
    parallel_mode = ParallelMode.STAND_ALONE
    context.set_auto_parallel_context(parallel_mode=parallel_mode, gradients_mean=True, device_num=1)

    args.logger.info('Creating Network....')
    network = YOLOV4CspDarkNet53(is_training=False)

    args.logger.info(args.pretrained)
    if os.path.isfile(args.pretrained):
        param_dict = load_checkpoint(args.pretrained)
        param_dict_new = {}
        for key, values in param_dict.items():
            if key.startswith('moments.'):
                continue
            elif key.startswith('yolo_network.'):
                param_dict_new[key[13:]] = values
            else:
                param_dict_new[key] = values
        load_param_into_net(network, param_dict_new)
        args.logger.info('load_model {} success'.format(args.pretrained))
    else:
        args.logger.info('{} not exists or not a pre-trained file'.format(args.pretrained))
        assert FileNotFoundError('{} not exists or not a pre-trained file'.format(args.pretrained))
        exit(1)

    data_root = args.data_root
    ann_file = args.ann_file

    config = ConfigYOLOV4CspDarkNet53()
    if args.testing_shape:
        config.test_img_shape = convert_testing_shape(args.testing_shape)

    ds, data_size = create_yolo_dataset(data_root, ann_file, is_training=False, batch_size=args.per_batch_size,
                                        max_epoch=1, device_num=1, rank=rank_id, shuffle=False,
                                        config=config)

    args.logger.info('testing shape : {}'.format(config.test_img_shape))
    args.logger.info('totol {} images to eval'.format(data_size))

    network.set_train(False)

    # init detection engine
    detection = DetectionEngine(args)

    input_shape = Tensor(tuple(config.test_img_shape), ms.float32)
    args.logger.info('Start inference....')
    for index, data in enumerate(ds.create_dict_iterator(num_epochs=1)):
        image = data["image"]

        image_shape_ = data["image_shape"]
        image_id_ = data["img_id"]

        prediction = network(image, input_shape)
        output_big, output_me, output_small = prediction
        output_big = output_big.asnumpy()
        output_me = output_me.asnumpy()
        output_small = output_small.asnumpy()
        image_id_ = image_id_.asnumpy()
        image_shape_ = image_shape_.asnumpy()

        detection.detect([output_small, output_me, output_big], args.per_batch_size, image_shape_, image_id_)
        if index % 1000 == 0:
            args.logger.info('Processing... {:.2f}% '.format(index * args.per_batch_size / data_size * 100))

    args.logger.info('Calculating mAP...')
    detection.do_nms_for_results()
    result_file_path = detection.write_result()
    args.logger.info('result file path: {}'.format(result_file_path))
    eval_result = detection.get_eval_result()

    cost_time = time.time() - start_time
    args.logger.info('\n=============coco eval reulst=========\n' + eval_result)
    args.logger.info('testing cost time {:.2f}h'.format(cost_time / 3600.))