upt/TestDETR.py at main · henlein/upt · GitHub

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
123
124
125
126
127
128
129
130
131
132
133
134
135
import torch
from transformers import DetrFeatureExtractor, DetrForObjectDetection
import configparser
from utils import custom_collate, CustomisedDLE, DataFactory
from torch import nn
from util import box_ops
from tqdm import tqdm
import json

def get_iou(bb1, bb2):
    assert bb1['x1'] < bb1['x2']
    assert bb1['y1'] < bb1['y2']
    assert bb2['x1'] < bb2['x2']
    assert bb2['y1'] < bb2['y2']

    # determine the coordinates of the intersection rectangle
    x_left = max(bb1['x1'], bb2['x1'])
    y_top = max(bb1['y1'], bb2['y1'])
    x_right = min(bb1['x2'], bb2['x2'])
    y_bottom = min(bb1['y2'], bb2['y2'])

    if x_right < x_left or y_bottom < y_top:
        return 0.0

    # The intersection of two axis-aligned bounding boxes is always an
    # axis-aligned bounding box
    intersection_area = (x_right - x_left) * (y_bottom - y_top)

    # compute the area of both AABBs
    bb1_area = (bb1['x2'] - bb1['x1']) * (bb1['y2'] - bb1['y1'])
    bb2_area = (bb2['x2'] - bb2['x1']) * (bb2['y2'] - bb2['y1'])

    # compute the intersection over union by taking the intersection
    # area and dividing it by the sum of prediction + ground-truth
    # areas - the interesection area
    iou = intersection_area / float(bb1_area + bb2_area - intersection_area)
    assert iou >= 0.0
    assert iou <= 1.0
    return iou


if __name__ == "__main__":
    configp = configparser.ConfigParser()
    configp.read('config.ini')

    threshold = 0.9
    #trainset = DataFactory(name="hicodet", partition="train2015", data_root="D:/Corpora/HICO-DET/")
    testset = DataFactory(name="hicodet", partition="test2015", data_root="D:/Corpora/HICO-DET/")

    model = DetrForObjectDetection.from_pretrained('facebook/detr-resnet-50')
    feature_extractor = DetrFeatureExtractor.from_pretrained('facebook/detr-resnet-50')

    correct = []
    wrong_label = []
    wrong_det = []
    not_det = []
    results = {}
    for idx, (img, data) in tqdm(enumerate(testset), total=len(testset)):
        if idx < 4:
            continue
        #print( testset.dataset.filename(idx))
        boxes_h = data["boxes_h"]
        boxes_o = data["boxes_o"]
        o_label = data["object"]
        target_sizes = data["size"]

        boxes_stack = torch.cat((boxes_h, boxes_o), 0)
        h_label = torch.ones(len(boxes_h))
        labels_stack = torch.cat((h_label, o_label), 0)
        print("==============================")
        print(torch.unsqueeze(img, 0))

        features, pos = model.model.backbone(torch.unsqueeze(img, 0))
        print(features)
        print("....")
        print(pos)
        exit()
        model_outputs = model.model(torch.unsqueeze(img, 0))[0]
        print("======")
        print(model_outputs)
        exit()
        prob = nn.functional.softmax(model_outputs.logits, -1)
        scores, labels = prob[..., :-1].max(-1)

        boxes = box_ops.box_cxcywh_to_xyxy(model_outputs["pred_boxes"])
        # and from relative [0, 1] to absolute [0, height] coordinates
        img_h, img_w = target_sizes.unbind(0)
        scale_fct = torch.stack([img_w, img_h, img_w, img_h], dim=0)
        boxes = boxes * scale_fct[None, :]


        results = [{'scores': s, 'labels': l, 'boxes': b} for s, l, b in zip(scores, labels, boxes)][0]

        keep = results["scores"] > threshold
        scores = results["scores"][keep]
        pred_labels = results["labels"][keep]
        pred_boxes = results["boxes"][keep]
        print(scores)
        print(pred_labels)
        print(pred_boxes)
        exit()
        boxes_stack = box_ops.box_cxcywh_to_xyxy(boxes_stack)
        boxes_stack = boxes_stack * scale_fct[None, :]

        correct.append(0)
        wrong_label.append(0)
        wrong_det.append(0)


        for pbox, plab in zip(pred_boxes, pred_labels):
            found = False
            for tbox, tlab in zip(boxes_stack, labels_stack):
                hbox_overlap = get_iou(
                    {"x1": pbox[0].item(), "x2": pbox[2].item(), "y1": pbox[1].item(), "y2": pbox[3].item()},
                    {"x1": tbox[0].item(), "x2": tbox[2].item(), "y1": tbox[1].item(), "y2": tbox[3].item()})

                if hbox_overlap > 0.5:
                    found = True
                    if plab == tlab:
                        correct[-1] += 1
                    else:
                        wrong_label[-1] += 1
                    break

            if not found:
                wrong_det[-1] += 1
        not_det.append(len(labels_stack) - sum([correct[-1], wrong_label[-1]]))
        print(str(correct[-1]), str(wrong_label[-1]), str(wrong_det[-1]), str(not_det[-1]))
        print(sum(correct), sum(wrong_label), sum(wrong_det), sum(not_det))
        print("===========")
        exit()
    print(sum(correct))
    print(sum(wrong_label))
    print(sum(wrong_det))
    print(sum(not_det))