def run(self):
super().run()
self.queue_stop_sign.clear() # 队列获取数据停止标志
features = None
need_post = False
frame_count = 0
while True:
if self.source_queue:
queue_item = self.source_queue.get()
print("CameraSingleEval --> get frame")
if queue_item.frame_index == QueueItem.PAUSE_SIGN_INDEX:
need_post = True
elif queue_item.frame_index == QueueItem.STOP_SIGN_INDEX:
need_post = True
self.queue_stop_sign.set()
else:
frame = transform(queue_item.frame)
self.transformed_video[frame_count] = frame
frame_count += 1
if frame_count == self.APP.Evaluate_Frame_Count:
need_post = True
if need_post and frame_count > 1:
print("CameraSingleEval --> 开始提取特征")
need_post = False
feature = get_features(self.transformed_video[0: frame_count, ],
frame_batch_size=self.APP.Get_Feature_Batch_Size,
device=self.APP.Device)
if features is None:
features = feature
else:
features = torch.cat((features, feature), 0)
frame_count = 0
if self.queue_stop_sign.isSet():
print("CameraSingleEval -- > 特征提取完毕，开始评估：")
features = torch.unsqueeze(features, 0) # torch.Size([1, len, xxxx])
with torch.no_grad():
input_length = features.shape[1] * torch.ones(1, 1)
outputs = self.model(features, input_length)
ans = outputs[0][0].to('cpu').numpy()
print("CameraSingleEval --> get ans = ", ans)
record_item = RecordItem(0, frame_count, ans)
record_item.set_method(Method.SINGLE_METHOD)
self.APP.result_records.append(record_item)
self.APP.app_refresh_eval_info()
messagebox.showinfo("整段评估", "视频质量为:" + str(ans))
self.APP.EVAL_END_SIGN.set()
self.APP.EVAL_STATE = State.FINISHED_STATE
print("CameraSingleEval stop ！")
break
def run(self):
super().run()
transformed_video = torch.zeros([self.APP.Evaluate_Frame_Count,
self.channel, self.height, self.width])
cur_frame_index = 0
frame_count = 0
next_put_index = randint(0, self.APP.Evaluate_Frame_Count) # 下次需上传下标
tail_out_index = next_put_index + self.APP.Evaluate_Frame_Count
while cur_frame_index < self.video_data.shape[0]:
if cur_frame_index == tail_out_index:
last_put_index = tail_out_index # 上次发送帧下标
rand_count = randint(0, self.APP.Evaluate_Frame_Count)
next_put_index = last_put_index + rand_count
tail_out_index = next_put_index + self.APP.Evaluate_Frame_Count
print("camera_thread --> get rand_count = ", rand_count)
if next_put_index <= cur_frame_index < tail_out_index:
frame = self.video_data[cur_frame_index]
frame = Image.fromarray(frame)
frame = transform(frame)
transformed_video[frame_count] = frame
frame_count += 1
# 满足个数 | 视频最后一帧
can_get_features = frame_count == self.APP.Evaluate_Frame_Count \
or cur_frame_index == self.video_data.shape[0] - 1
can_get_features = can_get_features and frame_count > 1
if can_get_features:
features = get_features(transformed_video[0: frame_count, ],
frame_batch_size=self.APP.Get_Feature_Batch_Size,
device=self.APP.Device)
features = torch.unsqueeze(features, 0) # torch.Size([1, len, xxxx])
with torch.no_grad():
input_length = features.shape[1] * torch.ones(1, 1)
outputs = self.model(features, input_length)
ans = outputs[0][0].to('cpu').numpy()
print("VideoMultiEval --> get ans = ", ans)
record_item = RecordItem(cur_frame_index - frame_count, frame_count, ans)
record_item.set_method(Method.MULTI_METHOD)
self.APP.result_records.append(record_item)
self.APP.app_refresh_eval_info()
frame_count = 0
cur_frame_index += 1
self.APP.result_records.append(RecordItem.END_SIGN)
self.APP.EVAL_END_SIGN.set()
self.APP.EVAL_STATE = State.FINISHED_STATE
messagebox.showinfo(title="分段评估", message="分段评估已完成！")
def run(self):
super().run()
print("CameraMultiEval --> run()")
frame_count = 0
need_handle = False
self.queue_stop_sign.clear()
while True:
if self.source_queue: # 判断非空
queue_item = self.source_queue.get()
print("camera_evaluate_thread --> get frame")
if queue_item.frame == QueueItem.PAUSE_SIGN_INDEX:
need_handle = True
elif queue_item.frame == QueueItem.STOP_SIGN_INDEX:
need_handle = True
self.queue_stop_sign.set()
else:
frame = transform(queue_item.frame) # 转换格式
self.transformed_video[frame_count] = frame
frame_count += 1
if frame_count == self.APP.Evaluate_Frame_Count:
need_handle = True
if need_handle and frame_count > 1:
print("camera evaluate thread --> 开始评估")
need_handle = False
features = get_features(self.transformed_video[0: frame_count, ],
frame_batch_size=self.APP.Get_Feature_Batch_Size,
device=self.APP.Device)
features = torch.unsqueeze(features, 0) # torch.Size([1, len, xxxx])
with torch.no_grad():
input_length = features.shape[1] * torch.ones(1, 1)
outputs = self.model(features, input_length)
ans = outputs[0][0].to('cpu').numpy()
print("camera evaluate thread --> get ans = ", ans)
record_item = RecordItem(queue_item.frame_index - frame_count, frame_count, ans)
record_item.set_method(Method.MULTI_METHOD)
self.APP.result_records.append(record_item)
self.APP.app_refresh_eval_info()
frame_count = 0
# 阻塞至此
if self.queue_stop_sign.isSet():
# self.APP.result_records = self.records
self.APP.result_records.append(RecordItem.END_SIGN)
self.APP.EVAL_END_SIGN.set()
self.APP.EVAL_STATE = State.FINISHED_STATE
messagebox.showinfo("分段评估", "拍摄模式分段评估已完成！")
print("camera evaluate thread stop ！")
break
def run(self):
super().run()
cur_frame_index = self.start_index
self.transformed_video = torch.zeros([self.length, self.channel, self.height, self.width])
frame_count = self.video_data.shape[0]
if self.end_index > frame_count or self.end_index == 0:
self.end_index = frame_count
while cur_frame_index < self.end_index:
frame = self.video_data[cur_frame_index]
frame = Image.fromarray(frame)
frame = transform(frame)
self.transformed_video[cur_frame_index] = frame
cur_frame_index += 1
features = get_features(self.transformed_video,
frame_batch_size=self.APP.Get_Feature_Batch_Size,
device=self.APP.Device)
features = torch.unsqueeze(features, 0)
with torch.no_grad():
input_length = features.shape[1] * torch.ones(1, 1)
outputs = self.model(features, input_length)
ans = outputs[0][0].to('cpu').numpy()
record_item = RecordItem(0, cur_frame_index, ans)
if self.is_specified:
record_item.set_method(Method.SPECIFIED_METHOD)
else:
record_item.set_method(Method.SINGLE_METHOD)
self.APP.result_records.append(record_item)
self.APP.app_refresh_eval_info()
print("VideoSingleEval --> 获取评测结果：", ans)
self.APP.EVAL_END_SIGN.set()
self.APP.EVAL_STATE = State.FINISHED_STATE
messagebox.showinfo(title="整段评估结果:", message="视频质量为: " + str(ans))
print("VideoSingleEval --> end")
transformed_video = torch.zeros([video_length, video_channel, video_height, video_width])
transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])
for frame_idx in range(video_length):
frame = video_data[frame_idx]
frame = Image.fromarray(frame)
frame = transform(frame)
transformed_video[frame_idx] = frame
print('Video length: {}'.format(transformed_video.shape[0]))
# feature extraction
features = get_features(transformed_video, frame_batch_size=args.frame_batch_size, device=device)
features = torch.unsqueeze(features, 0) # batch size 1
# quality prediction using VSFA
model = VSFA()
model.load_state_dict(torch.load(args.model_path)) #
model.to(device)
model.eval()
with torch.no_grad():
input_length = features.shape[1] * torch.ones(1, 1)
outputs = model(features, input_length)
y_pred = outputs[0][0].to('cpu').numpy()
print("Predicted quality: {}".format(y_pred))
end = time.time()