def generate_encoded_samples(self, data_dir, tmp_dir, dataset_split):
"""Generate samples of the encoded frames with possible extra data.
By default this function just encodes the numpy array returned as "frame"
from `self.generate_samples` into a PNG image. Override this function to
get other encodings on disk.
Args:
data_dir: final data directory. Typically only used in this method to copy
over user-supplied vocab files if there are extra fields needing them.
tmp_dir: temporary directory that you can use for downloading and scratch.
dataset_split: problem.DatasetSplit, which data split to generate samples
for (for example, training and evaluation).
Yields:
Sample: dict<str feature_name, feature value> which is in disk encoding.
Raises:
ValueError: if the frame has a different number of channels than required.
"""
for features in self.generate_samples(data_dir, tmp_dir,
dataset_split):
unencoded_frame = features.pop("frame")
height, width, channels = unencoded_frame.shape
if channels != self.num_channels:
raise ValueError(
"Generated frame has %d channels while the class "
"assumes %d channels." % (channels, self.num_channels))
if height != self.frame_height:
raise ValueError(
"Generated frame has height %d while the class "
"assumes height %d." % (height, self.frame_height))
if width != self.frame_width:
raise ValueError(
"Generated frame has width %d while the class "
"assumes width %d." % (width, self.frame_width))
encoded_frame = six.next(
image_utils.encode_images_as_png([unencoded_frame]))
features["image/encoded"] = [encoded_frame]
features["image/format"] = ["png"]
features["image/height"] = [height]
features["image/width"] = [width]
yield features
def encode_image_to_png(image):
encoded = six.next(image_utils.encode_images_as_png([image]))
return encoded
