File size: 5,078 Bytes
938e515
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
# Copyright (c) Facebook, Inc. and its affiliates.
import copy
import logging
import numpy as np
import time
from pycocotools.cocoeval import COCOeval

from detectron2 import _C

logger = logging.getLogger(__name__)


class COCOeval_opt(COCOeval):
    """
    This is a slightly modified version of the original COCO API, where the functions evaluateImg()
    and accumulate() are implemented in C++ to speedup evaluation
    """

    def evaluate(self):
        """
        Run per image evaluation on given images and store results in self.evalImgs_cpp, a
        datastructure that isn't readable from Python but is used by a c++ implementation of
        accumulate().  Unlike the original COCO PythonAPI, we don't populate the datastructure
        self.evalImgs because this datastructure is a computational bottleneck.
        :return: None
        """
        tic = time.time()

        p = self.params
        # add backward compatibility if useSegm is specified in params
        if p.useSegm is not None:
            p.iouType = "segm" if p.useSegm == 1 else "bbox"
        logger.info("Evaluate annotation type *{}*".format(p.iouType))
        p.imgIds = list(np.unique(p.imgIds))
        if p.useCats:
            p.catIds = list(np.unique(p.catIds))
        p.maxDets = sorted(p.maxDets)
        self.params = p

        self._prepare()  # bottleneck

        # loop through images, area range, max detection number
        catIds = p.catIds if p.useCats else [-1]

        if p.iouType == "segm" or p.iouType == "bbox":
            computeIoU = self.computeIoU
        elif p.iouType == "keypoints":
            computeIoU = self.computeOks
        self.ious = {
            (imgId, catId): computeIoU(imgId, catId) for imgId in p.imgIds for catId in catIds
        }  # bottleneck

        maxDet = p.maxDets[-1]

        # <<<< Beginning of code differences with original COCO API
        def convert_instances_to_cpp(instances, is_det=False):
            # Convert annotations for a list of instances in an image to a format that's fast
            # to access in C++
            instances_cpp = []
            for instance in instances:
                instance_cpp = _C.InstanceAnnotation(
                    int(instance["id"]),
                    instance["score"] if is_det else instance.get("score", 0.0),
                    instance["area"],
                    bool(instance.get("iscrowd", 0)),
                    bool(instance.get("ignore", 0)),
                )
                instances_cpp.append(instance_cpp)
            return instances_cpp

        # Convert GT annotations, detections, and IOUs to a format that's fast to access in C++
        ground_truth_instances = [
            [convert_instances_to_cpp(self._gts[imgId, catId]) for catId in p.catIds]
            for imgId in p.imgIds
        ]
        detected_instances = [
            [convert_instances_to_cpp(self._dts[imgId, catId], is_det=True) for catId in p.catIds]
            for imgId in p.imgIds
        ]
        ious = [[self.ious[imgId, catId] for catId in catIds] for imgId in p.imgIds]

        if not p.useCats:
            # For each image, flatten per-category lists into a single list
            ground_truth_instances = [[[o for c in i for o in c]] for i in ground_truth_instances]
            detected_instances = [[[o for c in i for o in c]] for i in detected_instances]

        # Call C++ implementation of self.evaluateImgs()
        self._evalImgs_cpp = _C.COCOevalEvaluateImages(
            p.areaRng, maxDet, p.iouThrs, ious, ground_truth_instances, detected_instances
        )
        self._evalImgs = None

        self._paramsEval = copy.deepcopy(self.params)
        toc = time.time()
        logger.info("COCOeval_opt.evaluate() finished in {:0.2f} seconds.".format(toc - tic))
        # >>>> End of code differences with original COCO API

    def accumulate(self):
        """
        Accumulate per image evaluation results and store the result in self.eval.  Does not
        support changing parameter settings from those used by self.evaluate()
        """
        logger.info("Accumulating evaluation results...")
        tic = time.time()
        assert hasattr(
            self, "_evalImgs_cpp"
        ), "evaluate() must be called before accmulate() is called."

        self.eval = _C.COCOevalAccumulate(self._paramsEval, self._evalImgs_cpp)

        # recall is num_iou_thresholds X num_categories X num_area_ranges X num_max_detections
        self.eval["recall"] = np.array(self.eval["recall"]).reshape(
            self.eval["counts"][:1] + self.eval["counts"][2:]
        )

        # precision and scores are num_iou_thresholds X num_recall_thresholds X num_categories X
        # num_area_ranges X num_max_detections
        self.eval["precision"] = np.array(self.eval["precision"]).reshape(self.eval["counts"])
        self.eval["scores"] = np.array(self.eval["scores"]).reshape(self.eval["counts"])
        toc = time.time()
        logger.info("COCOeval_opt.accumulate() finished in {:0.2f} seconds.".format(toc - tic))