opencvによる映像処理後のウィンドウが「応答なし」となって閉じれず、その後の処理が実行できません。

問題

映像から表情認識を行うプログラムを使いたいのですが、表情認識自体はできているものの、映像が最後までいくと、出力された映像のウィンドウが「応答なし」となり、その後の処理を実行できません。

映像の途中であれば、キーボードからの入力によりウィンドウを閉じることはできます。
映像が最後までいっても「応答なし」の状態にならずにその後の処理を実行できるようにするにはどうしたらよいのか教えてください。

下記に、プログラムを記載します。

Python
1import os
2os.environ["OMP_NUM_THREADS"] = "2"
3import glob
4import json
5import cv2
6import numpy as np
7import torch
8from torchvision.transforms import transforms
9from models import densenet121, resmasking_dropout1
10import time
11import pandas as pd
12
13
14
15def ensure_color(image):
16    if len(image.shape) == 2:
17        return np.dstack([image] * 3)
18    elif image.shape[2] == 1:
19        return np.dstack([image] * 3)
20    return image
21
22
23net = cv2.dnn.readNetFromCaffe(
24    "deploy.prototxt.txt", "res10_300x300_ssd_iter_140000.caffemodel"
25)
26
27
28transform = transforms.Compose([transforms.ToPILImage(), transforms.ToTensor()])
29
30
31FER_2013_EMO_DICT = {
32    0: "angry",
33    1: "disgust",
34    2: "fear",
35    3: "happy",
36    4: "sad",
37    5: "surprise",
38    6: "neutral",
39}
40
41#Time = []
42#emo = []
43
44
45def main():
46    # load configs and set random seed
47    configs = json.load(open("./configs/fer2013_config.json"))
48    image_size = (configs["image_size"], configs["image_size"])
49
50    # model = densenet121(in_channels=3, num_classes=7)
51    model = resmasking_dropout1(in_channels=3, num_classes=7)
52    model.cuda()
53
54    # state = torch.load('./saved/checkpoints/densenet121_rot30_2019Nov11_14.23')
55    # state = torch.load('./saved/checkpoints/resmasking_dropout1_rot30_2019Nov17_14.33')
56    state = torch.load(
57        "./saved/checkpoints/Z_resmasking_dropout1_rot30_2019Nov30_13.32"
58    )
59    model.load_state_dict(state["net"])
60    model.eval()
61
62    vid = cv2.VideoCapture('NWH2_exported.mp4')
63
64
65    # cv2.namedWindow('disp')
66    # cv2.resizeWindow('disp', width=800)
67
68    with torch.no_grad():
69        #start_time = time.time()
70        while True:
71            ret, frame = vid.read()
72            if frame is None or ret is not True:
73                continue
74
75            try:
76                frame = np.fliplr(frame).astype(np.uint8)
77                # frame += 50
78                h, w = frame.shape[:2]
79                gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
80                # gray = frame
81
82                blob = cv2.dnn.blobFromImage(
83                    cv2.resize(frame, (300, 300)),
84                    1.0,
85                    (300, 300),
86                    (104.0, 177.0, 123.0),
87                )
88                net.setInput(blob)
89                faces = net.forward()
90
91                for i in range(0, faces.shape[2]):
92                    confidence = faces[0, 0, i, 2]
93                    if confidence < 0.5:
94                        continue
95                    box = faces[0, 0, i, 3:7] * np.array([w, h, w, h])
96                    start_x, start_y, end_x, end_y = box.astype("int")
97
98                    # covnert to square images
99                    center_x, center_y = (start_x + end_x) // 2, (start_y + end_y) // 2
100                    square_length = ((end_x - start_x) + (end_y - start_y)) // 2 // 2
101
102                    square_length *= 1.1
103
104                    start_x = int(center_x - square_length)
105                    start_y = int(center_y - square_length)
106                    end_x = int(center_x + square_length)
107                    end_y = int(center_y + square_length)
108
109                    cv2.rectangle(
110                        frame, (start_x, start_y), (end_x, end_y), (179, 255, 179), 2
111                    )
112                    # cv2.rectangle(frame , (x, y), (x + w, y + h), (179, 255, 179), 2)
113
114                    # face = gray[y:y + h, x:x + w]
115                    face = gray[start_y:end_y, start_x:end_x]
116
117                    face = ensure_color(face)
118
119                    face = cv2.resize(face, image_size)
120                    face = transform(face).cuda()
121                    face = torch.unsqueeze(face, dim=0)
122
123                    output = torch.squeeze(model(face), 0)
124                    proba = torch.softmax(output, 0)
125                    """
126                    now_time = time.time()
127                    delta_time = now_time - start_time
128                    Time.append(delta_time)
129                    """
130                    # emo_idx = torch.argmax(proba, dim=0).item()
131                    emo_proba, emo_idx = torch.max(proba, dim=0)
132                    emo_idx = emo_idx.item()
133                    emo_proba = emo_proba.item()
134
135                    emo_label = FER_2013_EMO_DICT[emo_idx]
136                    #emo.append(emo_label)
137
138                    label_size, base_line = cv2.getTextSize(
139                        "{}: 000".format(emo_label), cv2.FONT_HERSHEY_SIMPLEX, 0.8, 2
140                    )
141
142                    cv2.rectangle(
143                        frame,
144                        (end_x, start_y + 1 - label_size[1]),
145                        (end_x + label_size[0], start_y + 1 + base_line),
146                        (223, 128, 255),
147                        cv2.FILLED,
148                    )
149                    cv2.putText(
150                        frame,
151                        "{} {}".format(emo_label, int(emo_proba * 100)),
152                        (end_x, start_y + 1),
153                        cv2.FONT_HERSHEY_SIMPLEX,
154                        0.8,
155                        (0, 0, 0),
156                        2,
157                    )
158
159                cv2.imshow("disp", frame)
160                # cv2.imshow('disp', np.concatenate((gray ), axis=1))
161                #print(emo_label,emo_proba)
162
163
164                if cv2.waitKey(1) & 0xFF == ord("q"):
165                    break
166
167            except:
168                continue
169        vid.release()
170        cv2.destroyAllWindows()
171
172"""
173    data = {"Time": Time,
174            "Emotion": emo
175            }
176
177    fixed_emo = pd.DataFrame(data)
178    fixed_emo.to_csv('demo_emotions_2.csv', index=False)
179"""
180if __name__ == "__main__":
181    main()
182