def Demo(opt):
if opt.mode == 'video':
if opt.modality == 'R':
video_loader = VideoLoader(opt.video_path).start()
(fourcc, fps, w, h) = video_loader.videoinfo()
#load recognition loader
print('Loading model...')
sys.stdout.flush()
Recognition= RecognitionLoader(video_loader, recognition_classes, opt.temporal_sample_length, step=opt.recognize_sample_step).start()
#data writer
if opt.save_video:
if not os.path.exists(opt.out_path):
os.makedirs(opt.out_path)
save_path = os.path.join(opt.out_path, 'VideoDemo.avi')
writer = DataWriter(Recognition, savepath=save_path, save_video=opt.save_video, vis=opt.show, fps=fps, frameSize=(w, h)).start()
print('output video has been saved in %s'%save_path)
writer.stop()
elif opt.modality == 'DR':
if opt.save_video:
if not os.path.exists(opt.out_path):
os.makedirs(opt.out_path)
out_path = os.path.join(opt.out_path, 'result.avi')
video_DR = Video_Detect_Recognition_Processor(opt.video_path, out_path, detect_classes, recognition_classes).start()
video_DR.read_detect()
if opt.mode == 'camera':
if opt.modality == 'R':
CameraRecognitionLoader = CameraRecognition(classes).start()
elif opt.modality== 'DR':
#Camera = CameraLoader().start()
#Detect = Detect_Recognition_Processor(classes).start()
#Recognize = RecognizeProcessor(Detect, classes).start()
DR = Read_Detect_Recognition_Processor(detect_classes, recognition_classes).start()
DR.read_detect()
def main():
outSize = (64, 64)
sp = False
video_loader = VideoLoader("kunkun_nmsl_Trim.mp4",
batchSize=32,
wh_rate=(1, 1),
inDim=outSize,
sp=sp)
(fourcc, fps, frameSize, shape_dst, ow, oh) = video_loader.videoinfo()
# fps = 10
# frameSize = outSize
print(fourcc, fps, frameSize, shape_dst, ow, oh)
num_frames = video_loader.videoLen()
save_path = 'out_video/kunkun_nmsl_2.avi'
fourcc = cv2.VideoWriter_fourcc(*'XVID')
stream = cv2.VideoWriter(save_path, fourcc, fps, frameSize)
assert stream.isOpened(), 'Cannot open video for writing'
video_loader.start()
save_loader = saveFrame(video_loader, sp=sp).start()
ascend_loader = AscendHost(save_loader, sp=sp).start()
collect_loader = DataCollector(ascend_loader, num_job=8, sp=sp).start()
post_processor = PostProcessor(collect_loader,
remap=(shape_dst, ow, oh),
vis=True,
nobg=True,
sp=sp).start()
# post_processor = PostProcessor(collect_loader, remap=(shape_dst, ow, oh), vis=True, sp=sp).start()
cnt = 0
while True:
cnt += 1
start = time.time()
(img_render, img, orig_img, im_name) = post_processor.getitem()
if orig_img is None:
print('All done!')
break
# orig_img = img_render
cv2.imshow("Demo", orig_img)
cv2.waitKey(90)
stream.write(orig_img)
interval = time.time() - start
print("Time: {:.4f}\t{:04d}/{:04d}".format(interval, cnt, num_frames))
# sys.stdout.flush()
stream.release()
def get_det_processor(file_name):
videofile = file_name
mode = args.mode
if not os.path.exists(args.outputpath):
os.mkdir(args.outputpath)
if not len(videofile):
raise IOError('Error: must contain --video')
# Load input video
data_loader = VideoLoader(videofile, batchSize=args.detbatch).start()
(fourcc, fps, frameSize) = data_loader.videoinfo()
# Load detection loader
print('Loading YOLO model..')
sys.stdout.flush()
det_loader = DetectionLoader(data_loader, batchSize=args.detbatch).start()
det_processor = DetectionProcessor(det_loader).start()
return det_processor
def __init__(self, videofile, mode='normal'):
self.videofile = videofile
self.mode = mode
self.data_loader = VideoLoader(self.videofile,
batchSize=args.detbatch).start()
(fourcc, fps, frameSize) = self.data_loader.videoinfo()
self.fourcc = fourcc
self.fps = fps
self.frameSize = frameSize
self.det_loader = DetectionLoader(self.data_loader,
batchSize=args.detbatch).start()
self.det_processor = DetectionProcessor(self.det_loader).start()
self.pose_dataset = Mscoco()
save_path = os.path.join(
args.outputpath, 'AlphaPose_' +
ntpath.basename(self.videofile).split('.')[0] + '.mp4')
self.writer = DataWriter(args.save_video, save_path,
cv2.VideoWriter_fourcc(*'DIVX'), self.fps,
self.frameSize).start()
self.results = list()
def handle_video(videofile):
args.video = videofile
videofile = args.video
mode = args.mode
if not len(videofile):
raise IOError('Error: must contain --video')
# Load input video
data_loader = VideoLoader(videofile, batchSize=args.detbatch).start()
(fourcc, fps, frameSize) = data_loader.videoinfo()
print('the video is {} f/s'.format(fps))
# Load detection loader
print('Loading YOLO model..')
sys.stdout.flush()
det_loader = DetectionLoader(data_loader, batchSize=args.detbatch).start()
# start a thread to read frames from the file video stream
det_processor = DetectionProcessor(det_loader).start()
# Load pose model
pose_dataset = Mscoco()
if args.fast_inference:
pose_model = InferenNet_fast(4 * 1 + 1, pose_dataset)
else:
pose_model = InferenNet(4 * 1 + 1, pose_dataset)
pose_model.cuda()
pose_model.eval()
runtime_profile = {'dt': [], 'pt': [], 'pn': []}
# Data writer
save_path = os.path.join(
args.outputpath,
'AlphaPose_' + ntpath.basename(videofile).split('.')[0] + '.avi')
writer = DataWriter(args.save_video, save_path,
cv2.VideoWriter_fourcc(*'XVID'), fps,
frameSize).start()
im_names_desc = tqdm(range(data_loader.length()))
batchSize = args.posebatch
for i in im_names_desc:
start_time = getTime()
with torch.no_grad():
(inps, orig_img, im_name, boxes, scores, pt1,
pt2) = det_processor.read()
if orig_img is None:
break
if boxes is None or boxes.nelement() == 0:
writer.save(None, None, None, None, None, orig_img,
im_name.split('/')[-1])
continue
ckpt_time, det_time = getTime(start_time)
runtime_profile['dt'].append(det_time)
# Pose Estimation
datalen = inps.size(0)
leftover = 0
if (datalen) % batchSize:
leftover = 1
num_batches = datalen // batchSize + leftover
hm = []
for j in range(num_batches):
inps_j = inps[j * batchSize:min((j + 1) *
batchSize, datalen)].cuda()
hm_j = pose_model(inps_j)
hm.append(hm_j)
hm = torch.cat(hm)
ckpt_time, pose_time = getTime(ckpt_time)
runtime_profile['pt'].append(pose_time)
hm = hm.cpu().data
writer.save(boxes, scores, hm, pt1, pt2, orig_img,
im_name.split('/')[-1])
ckpt_time, post_time = getTime(ckpt_time)
runtime_profile['pn'].append(post_time)
if args.profile:
# TQDM
im_names_desc.set_description(
'det time: {dt:.4f} | pose time: {pt:.4f} | post processing: {pn:.4f}'
.format(dt=np.mean(runtime_profile['dt']),
pt=np.mean(runtime_profile['pt']),
pn=np.mean(runtime_profile['pn'])))
if (args.save_img or args.save_video) and not args.vis_fast:
print(
'===========================> Rendering remaining images in the queue...'
)
print(
'===========================> If this step takes too long, you can enable the --vis_fast flag to use fast rendering (real-time).'
)
while (writer.running()):
pass
writer.stop()
final_result = writer.results()
# 获取第 0 个框的人
kpts = []
for i in range(len(final_result)):
try:
preds = final_result[i]['result']
# preds[i]['keypoints'] (17,2)
# preds[i]['kp_score'] (17,1)
# preds[i]['proposal_score'] (1)
# 选择 y 坐标最大的人 —— 用于打羽毛球视频
max_index = 0
min_index = 0
# max_y = np.mean(preds[0]['keypoints'].data.numpy()[:, 1])
min_x = np.mean(preds[0]['keypoints'].data.numpy()[:, 0])
max_x = np.mean(preds[0]['keypoints'].data.numpy()[:, 0])
for k in range(len(preds)):
# tmp_y = np.mean(preds[k]['keypoints'].data.numpy()[:, 1])
tmp_x = np.mean(preds[k]['keypoints'].data.numpy()[:, 0])
# if tmp_y > max_y:
if tmp_x < min_x:
min_index = k
# max_y = tmp_y
min_x = tmp_x
for k in range(len(preds)):
# tmp_y = np.mean(preds[k]['keypoints'].data.numpy()[:, 1])
tmp_x = np.mean(preds[k]['keypoints'].data.numpy()[:, 0])
# if tmp_y > max_y:
if tmp_x > max_x:
max_index = k
max_x = tmp_x
mid_index = 0
for k in range(len(preds)):
if k == max_index or k == min_index:
continue
mid_index = k
kpt = preds[mid_index]['keypoints']
# kpt = final_result[i]['result'][0]['keypoints']
kpts.append(kpt.data.numpy())
except:
# print(sys.exc_info())
print('error...')
filename = os.path.basename(args.video).split('.')[0]
name = filename + '.npz'
kpts = np.array(kpts).astype(np.float32)
# print('kpts npz save in ', name)
# np.savez_compressed(name, kpts=kpts)
return kpts
args.dataset = 'coco'
if not args.sp:
torch.multiprocessing.set_start_method('forkserver', force=True)
torch.multiprocessing.set_sharing_strategy('file_system')
if __name__ == "__main__":
videofile = args.video
mode = args.mode
if not os.path.exists(args.outputpath):
os.mkdir(args.outputpath)
if not len(videofile):
raise IOError('Error: must contain --video')
# Load input video
data_loader = VideoLoader(videofile, batchSize=args.detbatch).start()
(fourcc, fps, frameSize) = data_loader.videoinfo()
# Load detection loader
print('Loading YOLO model..')
sys.stdout.flush()
det_loader = DetectionLoader(data_loader, batchSize=args.detbatch).start()
det_processor = DetectionProcessor(det_loader).start()
# Load pose model
pose_dataset = Mscoco()
if args.fast_inference:
pose_model = InferenNet_fast(4 * 1 + 1, pose_dataset)
else:
pose_model = InferenNet(4 * 1 + 1, pose_dataset)
pose_model.cuda()
def handle_video(video_file):
# =========== common ===============
args.video = video_file
base_name = os.path.basename(args.video)
video_name = base_name[:base_name.rfind('.')]
# =========== end common ===============
# =========== image ===============
# img_path = f'outputs/alpha_pose_{video_name}/split_image/'
# args.inputpath = img_path
# args.outputpath = f'outputs/alpha_pose_{video_name}'
# if os.path.exists(args.outputpath):
# shutil.rmtree(f'{args.outputpath}/vis', ignore_errors=True)
# else:
# os.mkdir(args.outputpath)
# # if not len(video_file):
# # raise IOError('Error: must contain --video')
# if len(img_path) and img_path != '/':
# for root, dirs, files in os.walk(img_path):
# im_names = sorted([f for f in files if 'png' in f or 'jpg' in f])
# else:
# raise IOError('Error: must contain either --indir/--list')
# # Load input images
# data_loader = ImageLoader(im_names, batchSize=args.detbatch, format='yolo').start()
# print(f'Totally {data_loader.datalen} images')
# =========== end image ===============
# =========== video ===============
args.outputpath = f'outputs/alpha_pose_{video_name}'
if os.path.exists(args.outputpath):
shutil.rmtree(f'{args.outputpath}/vis', ignore_errors=True)
else:
os.mkdir(args.outputpath)
videofile = args.video
mode = args.mode
if not len(videofile):
raise IOError('Error: must contain --video')
# Load input video
data_loader = VideoLoader(videofile, batchSize=args.detbatch).start()
(fourcc, fps, frameSize) = data_loader.videoinfo()
print('the video is {} f/s'.format(fps))
# =========== end video ===============
# Load detection loader
print('Loading YOLO model..')
sys.stdout.flush()
det_loader = DetectionLoader(data_loader, batchSize=args.detbatch).start()
# start a thread to read frames from the file video stream
det_processor = DetectionProcessor(det_loader).start()
# Load pose model
pose_dataset = Mscoco()
if args.fast_inference:
pose_model = InferenNet_fast(4 * 1 + 1, pose_dataset)
else:
pose_model = InferenNet(4 * 1 + 1, pose_dataset)
pose_model #.cuda()
pose_model.eval()
runtime_profile = {'dt': [], 'pt': [], 'pn': []}
# Data writer
save_path = os.path.join(
args.outputpath,
'AlphaPose_' + ntpath.basename(video_file).split('.')[0] + '.avi')
# writer = DataWriter(args.save_video, save_path, cv2.VideoWriter_fourcc(*'XVID'), fps, frameSize).start()
writer = DataWriter(args.save_video).start()
print('Start pose estimation...')
im_names_desc = tqdm(range(data_loader.length()))
batchSize = args.posebatch
for i in im_names_desc:
start_time = getTime()
with torch.no_grad():
(inps, orig_img, im_name, boxes, scores, pt1,
pt2) = det_processor.read()
if orig_img is None:
print(f'{i}-th image read None: handle_video')
break
if boxes is None or boxes.nelement() == 0:
writer.save(None, None, None, None, None, orig_img,
im_name.split('/')[-1])
continue
ckpt_time, det_time = getTime(start_time)
runtime_profile['dt'].append(det_time)
# Pose Estimation
datalen = inps.size(0)
leftover = 0
if datalen % batchSize:
leftover = 1
num_batches = datalen // batchSize + leftover
hm = []
for j in range(num_batches):
inps_j = inps[j * batchSize:min((j + 1) *
batchSize, datalen)] #.cuda()
hm_j = pose_model(inps_j)
hm.append(hm_j)
hm = torch.cat(hm)
ckpt_time, pose_time = getTime(ckpt_time)
runtime_profile['pt'].append(pose_time)
hm = hm.cpu().data
writer.save(boxes, scores, hm, pt1, pt2, orig_img,
im_name.split('/')[-1])
ckpt_time, post_time = getTime(ckpt_time)
runtime_profile['pn'].append(post_time)
if args.profile:
# TQDM
im_names_desc.set_description(
'det time: {dt:.4f} | pose time: {pt:.4f} | post processing: {pn:.4f}'
.format(dt=np.mean(runtime_profile['dt']),
pt=np.mean(runtime_profile['pt']),
pn=np.mean(runtime_profile['pn'])))
if (args.save_img or args.save_video) and not args.vis_fast:
print(
'===========================> Rendering remaining images in the queue...'
)
print(
'===========================> If this step takes too long, you can enable the --vis_fast flag to use fast rendering (real-time).'
)
while writer.running():
pass
writer.stop()
final_result = writer.results()
write_json(final_result, args.outputpath)
return final_result, video_name
def alphapose_process_video(videofile, pose_result_handler, inference_steps=1):
if not len(videofile):
raise IOError("Error: must contain --video")
# Load input video
print(f"Opening video {videofile}")
data_loader = VideoLoader(videofile, batchSize=args.detbatch).start()
(fourcc, fps, frameSize) = data_loader.videoinfo()
# Load detection loader
print("Loading YOLO model..")
sys.stdout.flush()
det_loader = DetectionLoader(
data_loader,
batchSize=args.detbatch,
path=ALPHAPOSE_DIR,
inference_steps=inference_steps,
).start()
det_processor = DetectionProcessor(det_loader).start()
# Load pose model
pose_dataset = Mscoco()
if args.fast_inference:
pose_model = InferenNet_fast(4 * 1 + 1,
pose_dataset,
path=ALPHAPOSE_DIR)
else:
pose_model = InferenNet(4 * 1 + 1, pose_dataset, path=ALPHAPOSE_DIR)
pose_model.cuda()
pose_model.eval()
runtime_profile = {"dt": [], "pt": [], "pn": []}
# Data writer
args.save_video = args.video_savefile is not None
writer = DataWriter(
save_video=args.
save_video, # Note: DataWriter uses args.save_video internally as well
savepath=args.video_savefile,
fourcc=cv2.VideoWriter_fourcc(*"XVID"),
fps=fps,
frameSize=frameSize,
result_handler=pose_result_handler,
).start()
im_names_desc = tqdm(range(data_loader.length()))
batchSize = args.posebatch
for i in im_names_desc:
start_time = getTime()
with torch.no_grad():
(inps, orig_img, im_name, boxes, scores, pt1,
pt2) = det_processor.read()
if orig_img is None:
break
if boxes is None or boxes.nelement() == 0:
writer.save(None, None, None, None, None, orig_img,
im_name.split("/")[-1])
continue
ckpt_time, det_time = getTime(start_time)
runtime_profile["dt"].append(det_time)
# Pose Estimation
datalen = inps.size(0)
leftover = 0
if (datalen) % batchSize:
leftover = 1
num_batches = datalen // batchSize + leftover
hm = []
for j in range(num_batches):
inps_j = inps[j * batchSize:min((j + 1) *
batchSize, datalen)].cuda()
hm_j = pose_model(inps_j)
hm.append(hm_j)
hm = torch.cat(hm)
ckpt_time, pose_time = getTime(ckpt_time)
runtime_profile["pt"].append(pose_time)
hm = hm.cpu().data
writer.save(boxes, scores, hm, pt1, pt2, orig_img,
im_name.split("/")[-1])
ckpt_time, post_time = getTime(ckpt_time)
runtime_profile["pn"].append(post_time)
if args.profile:
# TQDM
im_names_desc.set_description(
"det time: {dt:.3f} | pose time: {pt:.2f} | post processing: {pn:.4f}"
.format(
dt=np.mean(runtime_profile["dt"]),
pt=np.mean(runtime_profile["pt"]),
pn=np.mean(runtime_profile["pn"]),
))
print("===========================> Finish Model Running.")
if (args.save_img or args.video_savefile) and not args.vis_fast:
print(
"===========================> Rendering remaining images in the queue..."
)
print(
"===========================> If this step takes too long, you can enable "
"the --vis_fast flag to use fast rendering (real-time).")
while writer.running():
pass
writer.stop()
return writer.results()
args.outputpath, ('%s' % datetime.today().strftime('%y%m%d_%H%M%S')))
#init_detector
# detector = GroundTruthDetections()
#init_tracker
# tracker = Sort(use_dlib=use_dlibTracker) # create instance of the SORT tracker
if not os.path.exists(args.outputpath):
os.mkdir(args.outputpath)
if not len(videofile):
raise IOError('Error: must contain --video')
# Load input video
fvs = VideoLoader(videofile).start()
(fourcc, fps, frameSize) = fvs.videoinfo()
# Data writer
save_path = os.path.join(
args.outputpath,
'AlphaPose_' + videofile.split('/')[-1].split('.')[0] + '.avi')
writer = DataWriter(args.save_video, save_path,
cv2.VideoWriter_fourcc(*'XVID'), fps,
frameSize).start()
# Load YOLO model
print('Loading YOLO model..')
det_model = Darknet("yolo/cfg/yolov3.cfg")
det_model.load_weights('models/yolo/yolov3.weights')
det_model.net_info['height'] = args.inp_dim
def main(file_name):
# videofile = args.video
videofile = file_name
mode = args.mode
if not os.path.exists(args.outputpath):
os.mkdir(args.outputpath)
if not len(videofile):
raise IOError('Error: must contain --video')
# Load input video
data_loader = VideoLoader(videofile, batchSize=args.detbatch).start()
(fourcc, fps, frameSize) = data_loader.videoinfo()
# Load detection loader
print('Loading YOLO model..')
sys.stdout.flush()
det_loader = DetectionLoader(data_loader, batchSize=args.detbatch).start()
det_processor = DetectionProcessor(det_loader).start()
# Load pose model
pose_dataset = Mscoco()
if args.fast_inference:
pose_model = InferenNet_fast(4 * 1 + 1, pose_dataset)
else:
pose_model = InferenNet(4 * 1 + 1, pose_dataset)
pose_model.cuda()
pose_model.eval()
runtime_profile = {'dt': [], 'pt': [], 'pn': []}
# Data writer
save_path = os.path.join(
args.outputpath,
'AlphaPose_' + ntpath.basename(videofile).split('.')[0] + '.avi')
writer = DataWriter(args.save_video, save_path,
cv2.VideoWriter_fourcc(*'XVID'), fps,
frameSize).start()
im_names_desc = tqdm(range(data_loader.length()))
batchSize = args.posebatch
for i in im_names_desc:
start_time = getTime()
with torch.no_grad():
(inps, orig_img, im_name, boxes, scores, pt1,
pt2) = det_processor.read()
if orig_img is None:
break
if boxes is None or boxes.nelement() == 0:
writer.save(None, None, None, None, None, orig_img,
im_name.split('/')[-1])
continue
ckpt_time, det_time = getTime(start_time)
runtime_profile['dt'].append(det_time)
# Pose Estimation
datalen = inps.size(0)
leftover = 0
if (datalen) % batchSize:
leftover = 1
num_batches = datalen // batchSize + leftover
hm = []
for j in range(num_batches):
inps_j = inps[j * batchSize:min((j + 1) *
batchSize, datalen)].cuda()
hm_j = pose_model(inps_j)
hm.append(hm_j)
hm = torch.cat(hm)
ckpt_time, pose_time = getTime(ckpt_time)
runtime_profile['pt'].append(pose_time)
hm = hm.cpu().data
import ipdb
ipdb.set_trace()
writer.save(boxes, scores, hm, pt1, pt2, orig_img,
im_name.split('/')[-1])
ckpt_time, post_time = getTime(ckpt_time)
runtime_profile['pn'].append(post_time)
if args.profile:
# TQDM
im_names_desc.set_description(
'det time: {dt:.4f} | pose time: {pt:.4f} | post processing: {pn:.4f}'
.format(dt=np.mean(runtime_profile['dt']),
pt=np.mean(runtime_profile['pt']),
pn=np.mean(runtime_profile['pn'])))
print('===========================> Finish Model Running.')
if (args.save_img or args.save_video) and not args.vis_fast:
print(
'===========================> Rendering remaining images in the queue...'
)
print(
'===========================> If this step takes too long, you can enable the --vis_fast flag to use fast rendering (real-time).'
)
while (writer.running()):
pass
writer.stop()
final_result = writer.results()
write_json(final_result, args.outputpath)
def handle_video(videofile, no_nan=True):
args.video = videofile
videofile = args.video
mode = args.mode
if not len(videofile):
raise IOError('Error: must contain --video')
# Load input video
data_loader = VideoLoader(videofile, batchSize=args.detbatch).start()
(fourcc, fps, frameSize) = data_loader.videoinfo()
cam_w = frameSize[0]
cam_h = frameSize[1]
print('the video is {} f/s'.format(fps))
# Load detection loader
print('Loading YOLO model..')
sys.stdout.flush()
det_loader = DetectionLoader(data_loader, batchSize=args.detbatch).start()
# start a thread to read frames from the file video stream
det_processor = DetectionProcessor(det_loader).start()
# Load pose model
pose_dataset = Mscoco()
if args.fast_inference:
pose_model = InferenNet_fast(4 * 1 + 1, pose_dataset)
else:
pose_model = InferenNet(4 * 1 + 1, pose_dataset)
pose_model.cuda()
pose_model.eval()
runtime_profile = {'dt': [], 'pt': [], 'pn': []}
# Data writer
save_path = os.path.join(
args.outputpath,
'AlphaPose_' + ntpath.basename(videofile).split('.')[0] + '.avi')
writer = DataWriter(args.save_video, save_path,
cv2.VideoWriter_fourcc(*'XVID'), fps,
frameSize).start()
im_names_desc = tqdm(range(data_loader.length()))
batchSize = args.posebatch
frames_w_pose = []
frame_cnt = 0
for i in im_names_desc:
start_time = getTime()
with torch.no_grad():
(inps, orig_img, im_name, boxes, scores, pt1,
pt2) = det_processor.read()
if orig_img is None:
break
frame_cnt += 1
if boxes is None or boxes.nelement() == 0:
writer.save(None, None, None, None, None, orig_img,
im_name.split('/')[-1])
continue
frames_w_pose.append(frame_cnt - 1)
ckpt_time, det_time = getTime(start_time)
runtime_profile['dt'].append(det_time)
# Pose Estimation
datalen = inps.size(0)
leftover = 0
if (datalen) % batchSize:
leftover = 1
num_batches = datalen // batchSize + leftover
hm = []
for j in range(num_batches):
inps_j = inps[j * batchSize:min((j + 1) *
batchSize, datalen)].cuda()
hm_j = pose_model(inps_j)
hm.append(hm_j)
hm = torch.cat(hm)
ckpt_time, pose_time = getTime(ckpt_time)
runtime_profile['pt'].append(pose_time)
hm = hm.cpu().data
writer.save(boxes, scores, hm, pt1, pt2, orig_img,
im_name.split('/')[-1])
ckpt_time, post_time = getTime(ckpt_time)
runtime_profile['pn'].append(post_time)
if args.profile:
# TQDM
im_names_desc.set_description(
'det time: {dt:.4f} | pose time: {pt:.4f} | post processing: {pn:.4f}'
.format(dt=np.mean(runtime_profile['dt']),
pt=np.mean(runtime_profile['pt']),
pn=np.mean(runtime_profile['pn'])))
if (args.save_img or args.save_video) and not args.vis_fast:
print(
'===========================> Rendering remaining images in the queue...'
)
print(
'===========================> If this step takes too long, you can enable the --vis_fast flag to use fast rendering (real-time).'
)
while (writer.running()):
pass
writer.stop()
final_result = writer.results()
kpts = []
if not no_nan:
for i in range(frame_cnt):
# initialize to NaN so we can interpolate later
kpts.append(np.full((17, 2), np.nan, dtype=np.float32))
for i in range(len(final_result)):
try:
kpt = final_result[i]['result'][0]['keypoints']
if not no_nan:
kpts[frames_w_pose[i]] = kpt.data.numpy()
else:
kpts.append(kpt.data.numpy())
except:
print('error...')
kpts = np.array(kpts).astype(np.float32)
#filename = os.path.basename(args.video).split('.')[0]
#name = filename + '.npz'
#print('kpts npz save in ', name)
#np.savez_compressed(name, kpts=kpts, fps=fps, cam_w=cam_w, cam_h=cam_h)
return kpts, fps, cam_w, cam_h
def main():
save_path = './video_cap'
video_name = 'webcam.avi'
fourcc= cv2.VideoWriter_fourcc(*'XVID')
fps = 24
input("Press Any Key to Open WebCam (press 'q' to stop): ")
cap = cv2.VideoCapture(0)
frameSize = (int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)),
int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)))
writer = cv2.VideoWriter(save_path+'/'+video_name, fourcc, fps, frameSize)
for i in range(42):
cap.read()
start_write = False
frame_cnt = 0
while True:
ret, frame = cap.read()
cv2.imshow("capture", frame)
if start_write:
writer.write(frame)
frame_cnt += 1
if cv2.waitKey(1) & 0xFF == ord('s'):
print('Start Rec!')
start_write = True
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()
print("Start Processing!")
outSize = (128,128)
sp = False
video_loader = VideoLoader(save_path+'/'+video_name, batchSize=32, wh_rate=(1,1), inDim=outSize, sp=sp, dataLen=frame_cnt)
(fourcc, fps, frameSize, shape_dst, ow, oh) = video_loader.videoinfo()
# fps = 10
print(fourcc, fps, frameSize, shape_dst, ow, oh, video_loader.videoLen())
num_frames = video_loader.videoLen()
out_path = 'out_video/'+ video_name
fourcc=cv2.VideoWriter_fourcc(*'XVID')
stream = cv2.VideoWriter(out_path, fourcc, fps, frameSize)
assert stream.isOpened(), 'Cannot open video for writing'
video_loader.start()
save_loader = saveFrame(video_loader, sp=sp).start()
ascend_loader = AscendHost(save_loader, sp=sp).start()
collect_loader = DataCollector(ascend_loader, num_job=8, sp=sp).start()
post_processor = PostProcessor(collect_loader, remap=(shape_dst, ow, oh), vis=True, nobg=True, sp=sp).start()
# post_processor = PostProcessor(collect_loader, remap=(shape_dst, ow, oh), vis=True, sp=sp).start()
cnt = 0
while True:
cnt += 1
start = time.time()
(img_render, img, orig_img, im_name) = post_processor.getitem()
if orig_img is None:
print('All done!')
break
orig_img = img_render
cv2.imshow("Demo", orig_img)
cv2.waitKey(160)
stream.write(orig_img)
interval = time.time() - start
print("Time: {:.4f}\t{:04d}/{:04d}".format(interval, cnt, num_frames))
# sys.stdout.flush()
stream.release()