def predict_naive(model_dir,
video_file,
image_dir,
device='gpu',
threshold=0.5,
output_dir='output'):
pred_config = PredictConfig(model_dir)
detector = JDE_Detector(pred_config, model_dir, device=device.upper())
if video_file is not None:
predict_video(detector,
video_file,
threshold=threshold,
output_dir=output_dir,
save_images=True,
save_mot_txts=True,
draw_center_traj=False,
secs_interval=10,
do_entrance_counting=False)
else:
img_list = get_test_images(image_dir, infer_img=None)
predict_image(detector,
img_list,
threshold=threshold,
output_dir=output_dir,
save_images=True)
def main():
pred_config = PredictConfig(FLAGS.model_dir)
detector = JDE_Detector(pred_config,
FLAGS.model_dir,
device=FLAGS.device,
run_mode=FLAGS.run_mode,
trt_min_shape=FLAGS.trt_min_shape,
trt_max_shape=FLAGS.trt_max_shape,
trt_opt_shape=FLAGS.trt_opt_shape,
trt_calib_mode=FLAGS.trt_calib_mode,
cpu_threads=FLAGS.cpu_threads,
enable_mkldnn=FLAGS.enable_mkldnn)
# predict from video file or camera video stream
if FLAGS.video_file is not None or FLAGS.camera_id != -1:
predict_video(detector,
FLAGS.video_file,
threshold=FLAGS.threshold,
output_dir=FLAGS.output_dir,
save_images=FLAGS.save_images,
save_mot_txts=FLAGS.save_mot_txts,
draw_center_traj=FLAGS.draw_center_traj,
secs_interval=FLAGS.secs_interval,
do_entrance_counting=FLAGS.do_entrance_counting,
camera_id=FLAGS.camera_id)
else:
# predict from image
img_list = get_test_images(FLAGS.image_dir, FLAGS.image_file)
predict_image(detector,
img_list,
threshold=FLAGS.threshold,
output_dir=FLAGS.output_dir,
save_images=FLAGS.save_images,
run_benchmark=FLAGS.run_benchmark)
if not FLAGS.run_benchmark:
detector.det_times.info(average=True)
else:
mems = {
'cpu_rss_mb': detector.cpu_mem / len(img_list),
'gpu_rss_mb': detector.gpu_mem / len(img_list),
'gpu_util': detector.gpu_util * 100 / len(img_list)
}
perf_info = detector.det_times.report(average=True)
model_dir = FLAGS.model_dir
mode = FLAGS.run_mode
model_info = {
'model_name': model_dir.strip('/').split('/')[-1],
'precision': mode.split('_')[-1]
}
data_info = {
'batch_size': 1,
'shape': "dynamic_shape",
'data_num': perf_info['img_num']
}
det_log = PaddleInferBenchmark(detector.config, model_info,
data_info, perf_info, mems)
det_log('MOT')
def main():
deploy_file = os.path.join(FLAGS.model_dir, 'infer_cfg.yml')
with open(deploy_file) as f:
yml_conf = yaml.safe_load(f)
arch = yml_conf['arch']
detector = SDE_Detector(
FLAGS.model_dir,
tracker_config=FLAGS.tracker_config,
device=FLAGS.device,
run_mode=FLAGS.run_mode,
batch_size=1,
trt_min_shape=FLAGS.trt_min_shape,
trt_max_shape=FLAGS.trt_max_shape,
trt_opt_shape=FLAGS.trt_opt_shape,
trt_calib_mode=FLAGS.trt_calib_mode,
cpu_threads=FLAGS.cpu_threads,
enable_mkldnn=FLAGS.enable_mkldnn,
output_dir=FLAGS.output_dir,
threshold=FLAGS.threshold,
save_images=FLAGS.save_images,
save_mot_txts=FLAGS.save_mot_txts,
draw_center_traj=FLAGS.draw_center_traj,
secs_interval=FLAGS.secs_interval,
do_entrance_counting=FLAGS.do_entrance_counting,
reid_model_dir=FLAGS.reid_model_dir,
mtmct_dir=FLAGS.mtmct_dir,
)
# predict from video file or camera video stream
if FLAGS.video_file is not None or FLAGS.camera_id != -1:
detector.predict_video(FLAGS.video_file, FLAGS.camera_id)
elif FLAGS.mtmct_dir is not None:
with open(FLAGS.mtmct_cfg) as f:
mtmct_cfg = yaml.safe_load(f)
detector.predict_mtmct(FLAGS.mtmct_dir, mtmct_cfg)
else:
# predict from image
if FLAGS.image_dir is None and FLAGS.image_file is not None:
assert FLAGS.batch_size == 1, "--batch_size should be 1 in MOT models."
img_list = get_test_images(FLAGS.image_dir, FLAGS.image_file)
seq_name = FLAGS.image_dir.split('/')[-1]
detector.predict_image(img_list,
FLAGS.run_benchmark,
repeats=10,
seq_name=seq_name)
if not FLAGS.run_benchmark:
detector.det_times.info(average=True)
else:
mode = FLAGS.run_mode
model_dir = FLAGS.model_dir
model_info = {
'model_name': model_dir.strip('/').split('/')[-1],
'precision': mode.split('_')[-1]
}
bench_log(detector, img_list, model_info, name='MOT')
def predict_naive(model_dir,
reid_model_dir,
video_file,
image_dir,
mtmct_dir=None,
mtmct_cfg=None,
scaled=True,
device='gpu',
threshold=0.5,
output_dir='output'):
pred_config = PredictConfig(model_dir)
detector_func = 'SDE_Detector'
if pred_config.arch == 'PicoDet':
detector_func = 'SDE_DetectorPicoDet'
detector = eval(detector_func)(pred_config, model_dir, device=device)
pred_config = PredictConfig(reid_model_dir)
reid_model = SDE_ReID(pred_config, reid_model_dir, device=device)
if video_file is not None:
predict_video(
detector,
reid_model,
video_file,
scaled=scaled,
threshold=threshold,
output_dir=output_dir,
save_images=True,
save_mot_txts=True,
draw_center_traj=False,
secs_interval=10,
do_entrance_counting=False)
elif mtmct_dir is not None:
with open(mtmct_cfg) as f:
mtmct_cfg_file = yaml.safe_load(f)
predict_mtmct(
detector,
reid_model,
mtmct_dir,
mtmct_cfg_file,
scaled=scaled,
threshold=threshold,
output_dir=output_dir,
save_images=True,
save_mot_txts=True)
else:
img_list = get_test_images(image_dir, infer_img=None)
predict_image(
detector,
reid_model,
img_list,
threshold=threshold,
output_dir=output_dir,
save_images=True)
def main():
detector = JDE_Detector(
FLAGS.model_dir,
tracker_config=None,
device=FLAGS.device,
run_mode=FLAGS.run_mode,
batch_size=1,
trt_min_shape=FLAGS.trt_min_shape,
trt_max_shape=FLAGS.trt_max_shape,
trt_opt_shape=FLAGS.trt_opt_shape,
trt_calib_mode=FLAGS.trt_calib_mode,
cpu_threads=FLAGS.cpu_threads,
enable_mkldnn=FLAGS.enable_mkldnn,
output_dir=FLAGS.output_dir,
threshold=FLAGS.threshold,
save_images=FLAGS.save_images,
save_mot_txts=FLAGS.save_mot_txts,
draw_center_traj=FLAGS.draw_center_traj,
secs_interval=FLAGS.secs_interval,
do_entrance_counting=FLAGS.do_entrance_counting,
)
# predict from video file or camera video stream
if FLAGS.video_file is not None or FLAGS.camera_id != -1:
detector.predict_video(FLAGS.video_file, FLAGS.camera_id)
else:
# predict from image
img_list = get_test_images(FLAGS.image_dir, FLAGS.image_file)
detector.predict_image(img_list, FLAGS.run_benchmark, repeats=10)
if not FLAGS.run_benchmark:
detector.det_times.info(average=True)
else:
mode = FLAGS.run_mode
model_dir = FLAGS.model_dir
model_info = {
'model_name': model_dir.strip('/').split('/')[-1],
'precision': mode.split('_')[-1]
}
bench_log(detector, img_list, model_info, name='MOT')
def main():
pred_config = PredictConfig(FLAGS.model_dir)
detector_func = 'SDE_Detector'
if pred_config.arch == 'PicoDet':
detector_func = 'SDE_DetectorPicoDet'
detector = eval(detector_func)(pred_config,
FLAGS.model_dir,
device=FLAGS.device,
run_mode=FLAGS.run_mode,
batch_size=FLAGS.batch_size,
trt_min_shape=FLAGS.trt_min_shape,
trt_max_shape=FLAGS.trt_max_shape,
trt_opt_shape=FLAGS.trt_opt_shape,
trt_calib_mode=FLAGS.trt_calib_mode,
cpu_threads=FLAGS.cpu_threads,
enable_mkldnn=FLAGS.enable_mkldnn)
pred_config = PredictConfig(FLAGS.reid_model_dir)
reid_model = SDE_ReID(pred_config,
FLAGS.reid_model_dir,
device=FLAGS.device,
run_mode=FLAGS.run_mode,
batch_size=FLAGS.reid_batch_size,
trt_min_shape=FLAGS.trt_min_shape,
trt_max_shape=FLAGS.trt_max_shape,
trt_opt_shape=FLAGS.trt_opt_shape,
trt_calib_mode=FLAGS.trt_calib_mode,
cpu_threads=FLAGS.cpu_threads,
enable_mkldnn=FLAGS.enable_mkldnn)
# predict from video file or camera video stream
if FLAGS.video_file is not None or FLAGS.camera_id != -1:
predict_video(detector,
reid_model,
FLAGS.video_file,
scaled=FLAGS.scaled,
threshold=FLAGS.threshold,
output_dir=FLAGS.output_dir,
save_images=FLAGS.save_images,
save_mot_txts=FLAGS.save_mot_txts,
draw_center_traj=FLAGS.draw_center_traj,
secs_interval=FLAGS.secs_interval,
do_entrance_counting=FLAGS.do_entrance_counting,
camera_id=FLAGS.camera_id)
elif FLAGS.mtmct_dir is not None:
mtmct_cfg_file = FLAGS.mtmct_cfg
with open(mtmct_cfg_file) as f:
mtmct_cfg = yaml.safe_load(f)
predict_mtmct(detector,
reid_model,
FLAGS.mtmct_dir,
mtmct_cfg,
scaled=FLAGS.scaled,
threshold=FLAGS.threshold,
output_dir=FLAGS.output_dir,
save_images=FLAGS.save_images,
save_mot_txts=FLAGS.save_mot_txts)
else:
# predict from image
img_list = get_test_images(FLAGS.image_dir, FLAGS.image_file)
predict_image(detector,
reid_model,
img_list,
threshold=FLAGS.threshold,
output_dir=FLAGS.output_dir,
save_images=FLAGS.save_images,
run_benchmark=FLAGS.run_benchmark)
if not FLAGS.run_benchmark:
detector.det_times.info(average=True)
reid_model.det_times.info(average=True)
else:
mode = FLAGS.run_mode
det_model_dir = FLAGS.model_dir
det_model_info = {
'model_name': det_model_dir.strip('/').split('/')[-1],
'precision': mode.split('_')[-1]
}
bench_log(detector, img_list, det_model_info, name='Det')
reid_model_dir = FLAGS.reid_model_dir
reid_model_info = {
'model_name': reid_model_dir.strip('/').split('/')[-1],
'precision': mode.split('_')[-1]
}
bench_log(reid_model, img_list, reid_model_info, name='ReID')