def __getitem__(self, item) -> dict:
is_negative_pair = (self._state["rng"].rand() <
self._hyper_params["negative_pair_ratio"])
data1 = data2 = None
while self.filt(data1) or self.filt(data2):
if is_negative_pair:
data1 = self._sample_track_frame()
data2 = self._sample_track_frame()
else:
data1, data2 = self._sample_track_pair()
data1["image"] = load_image(data1["image"])
data2["image"] = load_image(data2["image"])
sampled_data = dict(
data1=data1,
data2=data2,
is_negative_pair=is_negative_pair,
)
return sampled_data
from videoanalyst.utils import load_image
datasets = dataset_buidler.build(
task, task_cfg.data.sampler.submodules.dataset)
dataset = datasets[0]
while True:
# pick a frame randomly
seq_idx = np.random.choice(range(len(dataset)))
seq_idx = int(seq_idx)
seq = dataset[seq_idx]
# video dataset
if len(seq['image']) > 1:
frame_idx = np.random.choice(range(len(seq['image'])))
frame = {k: seq[k][frame_idx] for k in seq}
# fetch & visualize data
im = load_image(frame['image'])
anno = frame['anno']
# static image dataset
else:
im = load_image(seq['image'][0])
num_anno = len(seq['anno'])
if num_anno <= 0:
logger.info("no annotation")
continue
anno_idx = np.random.choice(num_anno)
anno = seq['anno'][anno_idx]
cv2.rectangle(im,
tuple(map(int, anno[:2])),
tuple(map(int, anno[2:])), (0, 255, 0),
thickness=3)
im = cv2.resize(im, (0, 0), fx=0.33, fy=0.33)
datasets = dataset_buidler.build(
task, task_cfg.data.sampler.submodules.dataset)
dataset = datasets[0]
while True:
# pick a frame randomly
seq_idx = np.random.choice(range(len(dataset)))
seq_idx = int(seq_idx)
seq = dataset[seq_idx]
# for one video
if len(seq['image']) > 1:
frame_idx = np.random.choice(range(len(seq['image'])))
frame = {k: seq[k][frame_idx] for k in seq}
else:
frame = dict(image=seq['image'][0], anno=seq['anno'])
# fetch & visualize data
im = load_image(frame['image'])
if parsed_args.data_type == "bbox":
cv2.rectangle(im,
tuple(map(int, frame['anno'][:2])),
tuple(map(int, frame['anno'][2:])), (0, 255, 0),
thickness=3)
im = cv2.resize(im, (0, 0), fx=0.33, fy=0.33)
cv2.imshow("im", im)
cv2.waitKey(0)
elif parsed_args.data_type == "mask":
cv2.imshow("im", im)
print(frame['anno'][0])
mask = (frame['anno'][0] * 50).astype(np.uint8).copy()
cv2.imwrite("mask_0.png", mask)
cv2.waitKey(0)
else: