def __getitem__(self, idx):
video = self.file_list[idx]
frames = []
masks = []
opt_flows = []
n_objects = []
frame_indexes = self._get_frame_indexes(video['n_frames'],
self.n_max_frames)
for fi in frame_indexes:
frame = np.array(IO.get(video['frames'][fi]).convert('RGB'))
frames.append(np.array(frame))
mask = IO.get(video['masks'][fi])
mask = mask.convert('P') if mask is not None else np.zeros(
frame.shape[:-1])
masks.append(np.array(mask))
opt_flow = IO.get(
video['optical_flow'][fi]) if 'optical_flow' in video else None
opt_flow = opt_flow if opt_flow is not None else np.zeros(
frame.shape[:-1] + (2, ))
opt_flows.append(np.array(opt_flow))
# The number of objects in the masks
mask_indexes = set()
for m in masks:
_mask_indexes = np.unique(m)
_mask_indexes = _mask_indexes[_mask_indexes != self.ignore_idx]
mask_indexes.update(_mask_indexes)
_n_objects = min(len(mask_indexes) - 1, self.n_max_objects)
n_objects.append(_n_objects)
# Data preprocessing and augmentation
if self.transforms is not None:
frames, masks, opt_flows = self.transforms(frames, masks,
opt_flows)
return video['name'], np.array(n_objects), frames, masks, opt_flows
def __getitem__(self, idx):
sample = self.file_list[idx]
data = {}
rand_idx = -1
if 'n_renderings' in self.options:
rand_idx = random.randint(0, self.options['n_renderings'] - 1) if self.options['shuffle'] else 0
for ri in self.options['required_items']:
file_path = sample['%s_path' % ri]
if type(file_path) == list:
file_path = file_path[rand_idx]
# print(file_path)
data[ri] = IO.get(file_path).astype(np.float32)
if self.transforms is not None:
data = self.transforms(data)
return sample['taxonomy_id'], sample['model_id'], data
def __getitem__(self, idx):
sample = self.file_list[idx]
data = {}
# rand_idx = -1
# if 'n_renderings' in self.options:
rand_idx = random.randint(0, 8 - 1) if self.options['shuffle'] else 0
# print('rand_idx2: ', rand_idx)
for ri in self.options['required_items']: # 'partial' & 'gt'
file_path = sample['%s_path' % ri]
if type(file_path) == list:
file_path = file_path[rand_idx]
# print('rand_idx1: ', rand_idx)
# 返回值是三维坐标的列表(np)
data[ri] = IO.get(file_path).astype(np.float32)
# print('data\'s shape: ', data['partial_cloud'].shape)
if self.transforms is not None:
data = self.transforms(data)
return sample['taxonomy_id'], sample['model_id'], data
