def inference_no_model(
data_loader,
iou_types=("bbox", ),
motion_specific=False,
box_only=False,
expected_results=(),
expected_results_sigma_tol=4,
output_folder=None,
):
dataset = data_loader.dataset
predictions = torch.load(os.path.join(output_folder, "predictions.pth"))
print("prediction loaded.")
extra_args = dict(
box_only=box_only,
iou_types=iou_types,
motion_specific=motion_specific,
expected_results=expected_results,
expected_results_sigma_tol=expected_results_sigma_tol,
)
return evaluate(dataset=dataset,
predictions=predictions,
output_folder=output_folder,
**extra_args)
def inference(
cfg,
model,
data_loader,
dataset_name,
iou_types=("bbox", ),
box_only=False,
bbox_aug=False,
device="cuda",
expected_results=(),
expected_results_sigma_tol=4,
output_folder=None,
):
# convert to a torch.device for efficiency
device = torch.device(device)
num_devices = get_world_size()
logger = logging.getLogger("mega_core.inference")
dataset = data_loader.dataset
logger.info("Start evaluation on {} dataset({} images).".format(
dataset_name, len(dataset)))
total_timer = Timer()
inference_timer = Timer()
total_timer.tic()
predictions = compute_on_dataset(model, data_loader, device, bbox_aug,
cfg.MODEL.VID.METHOD, inference_timer)
# wait for all processes to complete before measuring the time
synchronize()
total_time = total_timer.toc()
total_time_str = get_time_str(total_time)
logger.info(
"Total run time: {} ({} s / img per device, on {} devices)".format(
total_time_str, total_time * num_devices / len(dataset),
num_devices))
total_infer_time = get_time_str(inference_timer.total_time)
logger.info(
"Model inference time: {} ({} s / img per device, on {} devices)".
format(
total_infer_time,
inference_timer.total_time * num_devices / len(dataset),
num_devices,
))
predictions = _accumulate_predictions_from_multiple_gpus(predictions)
if not is_main_process():
return
if output_folder:
torch.save(predictions, os.path.join(output_folder, "predictions.pth"))
extra_args = dict(
box_only=box_only,
iou_types=iou_types,
expected_results=expected_results,
expected_results_sigma_tol=expected_results_sigma_tol,
)
return evaluate(dataset=dataset,
predictions=predictions,
output_folder=output_folder,
**extra_args)
def inference(
cfg,
model,
data_loader,
dataset_name,
iou_types=("bbox", ),
motion_specific=False,
box_only=False,
bbox_aug=False,
device="cuda",
expected_results=(),
expected_results_sigma_tol=4,
output_folder=None,
):
# convert to a torch.device for efficiency
device = torch.device(device)
num_devices = get_world_size()
logger = logging.getLogger("mega_core.inference")
dataset = data_loader.dataset
logger.info("Start evaluation on {} dataset({} images).".format(
dataset_name, len(dataset)))
total_timer = Timer()
inference_timer = Timer()
total_timer.tic()
predictions = compute_on_dataset(model, data_loader, device, bbox_aug,
cfg.MODEL.VID.METHOD, inference_timer)
# wait for all processes to complete before measuring the time
synchronize()
total_time = total_timer.toc()
total_time_str = get_time_str(total_time)
logger.info(
"Total run time: {} ({} s / img per device, on {} devices)".format(
total_time_str, total_time * num_devices / len(dataset),
num_devices))
total_infer_time = get_time_str(inference_timer.total_time)
logger.info(
"Model inference time: {} ({} s / img per device, on {} devices)".
format(
total_infer_time,
inference_timer.total_time * num_devices / len(dataset),
num_devices,
))
predictions = _accumulate_predictions_from_multiple_gpus(predictions)
##########################################################################
## save result img as a video
# out_video='result.avi'
# fourcc = cv2.VideoWriter_fourcc(*'XVID')
# fgs=25
# size = (560, 220)
# video_writer = cv2.VideoWriter(out_video, fourcc, fgs, size)
outputset = output_folder.split('/')[-1]
if hasattr(predictions, "copy"):
score_thr = [0.5, 0.7, 0.9]
for _score in score_thr:
file_data = ""
with open(f'{outputset}_video_result_{_score}.txt', 'w') as f:
f.write(file_data)
predictionstmp = predictions.copy()
video_list = {}
alarmvideo_list = {}
frame_record = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
frame_num = 0
for image_id, prediction in enumerate(predictionstmp):
img, target, filename = dataset.get_visualization(image_id)
frame_id = int(filename.split("/")[-1])
## every 12 frames
# if frame_id%12 != 0:
# continue
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
img_info = dataset.get_img_info(image_id)
image_width = img_info["width"]
image_height = img_info["height"]
prediction = prediction.resize((image_width, image_height))
pred_bbox = prediction.bbox.numpy()
pred_label = prediction.get_field("labels").numpy()
pred_score = prediction.get_field("scores").numpy()
# keep = pred_score.gt(score_thr)
keep = pred_score > _score
boxes = pred_bbox[keep]
classes = pred_label[keep]
scores = pred_score[keep]
template = "{}: {:.2f}"
boxnum = 0
for box, score, label in zip(boxes, scores, classes):
if label == 1:
label = 'person'
boxnum += 1
elif label == 2:
label = 'head'
else:
label = 'unknown'
rect = [int(k) for k in box[:4]]
cv2.rectangle(img, (rect[0], rect[1]), (rect[2], rect[3]),
(0, 0, 255), 2)
s = template.format(label, score)
cv2.putText(img, s, (rect[0], rect[1]),
cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 0), 2)
video_dictname = filename.split(
'/')[-3] + '/' + filename.split('/')[-2]
if video_dictname not in video_list:
frame_record = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
frame_num = 0
video_result = {}
video_list[video_dictname] = video_result
alarmvideo_list[video_dictname] = 0
file_data_tmp = ""
for single_video in alarmvideo_list:
file_data_tmp += str(single_video) + ', value: ' + str(
alarmvideo_list[single_video]) + '\n'
with open(f'{outputset}_video_result_{_score}_tmp.txt',
'w') as f:
f.write(file_data_tmp)
#every 12 frames
# if alarmvideo_list[video_dictname] ==0 and boxnum>1:
# alarmvideo_list[video_dictname] =1
#every frame 8 warning
if boxnum > 1:
frame_record[frame_num % 10] = 1
else:
frame_record[frame_num % 10] = 0
if alarmvideo_list[video_dictname] == 0 and sum(
frame_record) > 7:
alarmvideo_list[video_dictname] = 1
frame_num += 1
if filename.split('/')[-1] not in video_result:
video_result[filename.split('/')[-1]] = boxnum
imgdir = os.path.join(output_folder, os.path.dirname(filename))
if not os.path.exists(imgdir):
os.makedirs(imgdir)
output_name = os.path.join(
imgdir, '{}_{}.jpg'.format(_score,
os.path.basename(filename)))
cv2.imwrite(output_name, img)
# video_writer.write(img)
#xml
create_tree(output_name)
for box, score, label in zip(boxes, scores, classes):
rect = [int(k) for k in box[:4]]
create_object(annotation, label, rect[0], rect[1], rect[2],
rect[3])
# for coordinate_list in coordinates_list:
# create_object(annotation, coordinate_list[0], coordinate_list[1], coordinate_list[2], coordinate_list[3])
# if coordinates_list==[]:
# break
# 将树模型写入xml文件
tree = ET.ElementTree(annotation)
# tree.write('%s.xml' % output_name.rstrip('.jpg'))
# tree = ET.ElementTree.parse('%s.xml' % output_name.rstrip('.jpg')) # 解析movies.xml这个文件
root = tree.getroot() # 得到根元素，Element类
pretty_xml(root, '\t', '\n') # 执行美化方法
tree.write('%s.xml' % output_name.rstrip('.jpg'))
# file_data=str(alarmvideo_list)
for single_video in alarmvideo_list:
file_data += str(single_video) + ', value: ' + str(
alarmvideo_list[single_video]) + '\n'
with open(f'{outputset}_video_result_{_score}.txt', 'a') as f:
f.write(file_data)
# video_writer.release()
##########################################################################
if not is_main_process():
return
if output_folder:
torch.save(predictions, os.path.join(output_folder, "predictions.pth"))
extra_args = dict(
box_only=box_only,
iou_types=iou_types,
motion_specific=motion_specific,
expected_results=expected_results,
expected_results_sigma_tol=expected_results_sigma_tol,
)
return evaluate(dataset=dataset,
predictions=predictions,
output_folder=output_folder,
**extra_args)