def to_batch_items(self, data_dict):
data_dict['images'] = torch.stack([
tensor_from_numpy_image(i, self.use_color_jitter)
for i in data_dict['images']
]) # S x C x H x W
if data_dict['images'].dim() == 3:
data_dict['images'].unsqueeze_(1)
data_dict['word_indices'] = torch.tensor(data_dict['word_indices'],
dtype=torch.long)
return data_dict
def to_batch_items(self, data_dict):
data_dict['images'] = torch.stack([
tensor_from_numpy_image(i, self.use_color_jitter)
for i in data_dict['images']
]) # S x C x H x W
if data_dict['images'].dim() == 3:
data_dict['images'].unsqueeze_(1)
data_dict['word_indices'] = torch.tensor(data_dict['word_indices'],
dtype=torch.long)
data_dict['actions'] = torch.tensor(data_dict['actions'],
dtype=torch.float32)
data_dict['controls'] = packb(
([str(c) for c in data_dict['controls']]))
data_dict['states'] = packb(([str(s) for s in data_dict['states']]))
return data_dict
def to_batch_items(self, data_dict):
data_dict['images'] = torch.stack([
tensor_from_numpy_image(i, self.use_color_jitter)
for i in data_dict['images']
]) # S x C x H x W
if data_dict['images'].dim() == 3:
data_dict['images'].unsqueeze_(1)
data_dict['actions'] = torch.tensor(data_dict['actions'],
dtype=torch.float32)
data_dict['type'] = fetch_name_from_road_option(data_dict['type'])
data_dict['controls'] = packb(
([str(c) for c in data_dict['controls']]))
data_dict['states'] = packb(([str(s) for s in data_dict['states']]))
data_dict['stops'] = torch.tensor(data_dict['stops'],
dtype=torch.float32).view(
data_dict['actions'].size(0), 1)
return data_dict
def _tensor_from_numpy_image(image: np.ndarray) -> torch.Tensor:
c = image.shape[2]
return tensor_from_numpy_image(image, False).view(1, 1, c, IMAGE_HEIGHT,
IMAGE_WIDTH)
