def generate_video(
video_option: List[str],
video_dir: Optional[str],
images: List[np.ndarray],
episode_id: Union[int, str],
checkpoint_idx: int,
metrics: Dict[str, float],
tb_writer: TensorboardWriter,
fps: int = 10,
) -> None:
r"""Generate video according to specified information.
Args:
video_option: string list of "tensorboard" or "disk" or both.
video_dir: path to target video directory.
images: list of images to be converted to video.
episode_id: episode id for video naming.
checkpoint_idx: checkpoint index for video naming.
metric_name: name of the performance metric, e.g. "spl".
metric_value: value of metric.
tb_writer: tensorboard writer object for uploading video.
fps: fps for generated video.
Returns:
None
"""
if len(images) < 1:
return
metric_strs = []
for k, v in metrics.items():
if isinstance(v, str):
metric_strs.append(f"{k}={v}")
else:
metric_strs.append(f"{k}={v:.2f}")
video_name = f"episode={episode_id}-ckpt={checkpoint_idx}-" + "-".join(
metric_strs)
if "disk" in video_option:
assert video_dir is not None
images_to_video(images, video_dir, video_name, fps=fps)
if "tensorboard" in video_option:
tb_writer.add_video_from_np_images(f"episode{episode_id}",
checkpoint_idx,
images,
fps=fps)
return video_name
def generate_video(
video_option: List[str],
video_dir: Optional[str],
images: List[np.ndarray],
episode_id: int,
checkpoint_idx: int,
tag: str,
metrics: Dict[str, float],
tb_writer: TensorboardWriter,
fps: int = 10,
) -> None:
r"""Generate video according to specified information.
Args:
video_option: string list of "tensorboard" or "disk" or both.
video_dir: path to target video directory.
images: list of images to be converted to video.
episode_id: episode id for video naming.
checkpoint_idx: checkpoint index for video naming.
info: metric dictionary
tag: Additional tag for naming video
tb_writer: tensorboard writer object for uploading video.
fps: fps for generated video.
Returns:
None
"""
print(len(images))
if len(images) < 1:
return
metric_strs = []
for k, v in metrics.items():
metric_strs.append(f"{k}={v:.2f}")
video_name = f"{tag}_episode={episode_id}-ckpt={checkpoint_idx}-" + "-".join(
metric_strs
)
if "disk" in video_option:
assert video_dir is not None
images_to_video(images, video_dir, video_name)
if "tensorboard" in video_option:
tb_writer.add_video_from_np_images(
f"episode{episode_id}", checkpoint_idx, images, fps=fps
)
def generate_video(
video_option: List[str],
video_dir: Optional[str],
images: List[np.ndarray],
episode_id: int,
checkpoint_idx: int,
metric_name: str,
metric_value: float,
tb_writer: TensorboardWriter,
fps: int = 10,
) -> None:
r"""Generate video according to specified information.
Args:
video_option: string list of "tensorboard" or "disk" or both.
video_dir: path to target video directory.
images: list of images to be converted to video.
episode_id: episode id for video naming.
checkpoint_idx: checkpoint index for video naming.
metric_name: name of the performance metric, e.g. "spl".
metric_value: value of metric.
tb_writer: tensorboard writer object for uploading video.
fps: fps for generated video.
Returns:
None
"""
if len(images) < 1:
return
video_name = f"episode{episode_id}_ckpt{checkpoint_idx}_{metric_name}{metric_value:.2f}"
if "disk" in video_option:
assert video_dir is not None
images_to_video(images, video_dir, video_name)
if "tensorboard" in video_option:
tb_writer.add_video_from_np_images(f"episode{episode_id}",
checkpoint_idx,
images,
fps=fps)
