SDK Core ML Utils API
Miscellaneous operations in Pytorch
- metavision_core_ml.utils.torch_ops.cuda_tick()
Measures time for torch operations on gpu.
- metavision_core_ml.utils.torch_ops.cuda_time(func)
Decorator for Pytorch ops on gpu.
- Parameters
func – method to time
- metavision_core_ml.utils.torch_ops.normalize_tiles(tensor, num_stds=6, num_dims=2, real_min_max=True)
Normalizes tiles, allows us to have normalized views (we filter outliers + standardize)
- Parameters
tensor – tensor input, we assume last 2 dims are H,W
- Returns
normalized tensor
- Return type
tensor
- metavision_core_ml.utils.torch_ops.viz_flow(flow)
Visualizes flow in rgb colors
- Parameters
flow – (B,2,H,W) tensor
- Returns
(B,3,H,W) tensor
- Return type
rgb
Tools common to training main functions.
- metavision_core_ml.utils.train_utils.search_latest_checkpoint(root_dir, mode='time')
looks for latest checkpoint in latest sub-directory
A collection of utilities for searching image/video files.
- metavision_core_ml.utils.files.grab_files(adir, extensions, recursive=True)
Grabs files with allowed extensions
- Parameters
adir (str) – directory
extensions (list) – allowed extensions
recursive (boolean) – search recursively
- Returns
files
- metavision_core_ml.utils.files.grab_h5s(adir, recursive=True)
Grabs h5 files
- Parameters
adir – directory with h5 files
recursive (boolean) – search recursively
- Returns
h5 files
- Return type
files
- metavision_core_ml.utils.files.grab_images(adir, recursive=True)
Grabs image files
- Parameters
adir – directory with images
- Returns
image files
- Return type
files
- metavision_core_ml.utils.files.grab_images_and_videos(adir, recursive=True)
Grabs image and video files
- Parameters
adir – directory with images
- Returns
image and video files
- Return type
files
- metavision_core_ml.utils.files.grab_jsons(adir, recursive=True)
Grabs json files
- Parameters
adir – directory with json files
recursive (boolean) – search recursively
- Returns
json files
- Return type
files
- metavision_core_ml.utils.files.grab_tiff_images(adir, recursive=True)
Grabs tiff image files
- Parameters
adir – directory with tiff images
- Returns
image files
- Return type
files
- metavision_core_ml.utils.files.grab_videos(adir, recursive=True)
Grabs videos in a directory
- Parameters
adir (str) – directory
recursive (boolean) – search recursively
- Returns
files with image/ video extension
- Return type
files
- metavision_core_ml.utils.files.is_image(path)
Checks if a path is an image
- Parameters
path – file path
- Returns
True or False
- Return type
is_image (bool)
- metavision_core_ml.utils.files.is_tiff_image(path)
Checks if a path is a tiff image
- Parameters
path – file path
- Returns
True or False
- Return type
is_image (bool)
- metavision_core_ml.utils.files.is_video(path)
Checks if a path is a video
- Parameters
path – file path
- Returns
True or False
- Return type
is_video (bool)
This wrapper shows and/or writes a video
- class metavision_core_ml.utils.show_or_write.ShowWrite(window_name, video_path, crf=30)
ShowWrite either shows and/or writes a video
- Parameters
window_name (str) – if empty, will not show
video_path (str) – if empty, will not write
crf (int) – compression factor for video output