lightning_pose.utils

lightning_pose.utils.cropzoom Module

Tools for cropping labeled frames and videos to bounding-box regions of interest.

`generate_cropped_labeled_frames`(...)	Given model predictions, generates a bbox.csv, crops frames, and a cropped csv file.
`generate_cropped_video`(input_video_file, ...)	TODO make consistent with generate_cropped_labeled_frames
`generate_cropped_csv_file`(input_csv_file, ...)	Adjusts coordinates in the input CSV file either by adding or subtracting corresponding values from a bounding box CSV file.

Path handling functions.

`ckpt_path_from_base_path`(base_path, model_name)	Given a path to a hydra output with trained model, extract the model .ckpt file.
`get_keypoint_names`([cfg, csv_file, header_rows])	Return keypoint names from a label CSV file or from the config.
`return_absolute_path`(possibly_relative_path)	Return absolute path from possibly relative path.
`return_absolute_data_paths`(data_cfg[, ...])	Generate absolute path for our example toy data.
`extract_session_name_from_video`(...)	Extract session name from video filename by removing the view name.
`find_video_files_for_views`(video_dir, view_names)	Search inside a folder to find a list of videos from different sessions and views, split them up and return a list of lists like [[sess0_view0.mp4, sess0_view1.mp4, ...], [sess1_view0.mp4, sess1_view1.mp4, ...], ...]
`get_videos_in_dir`()	Gather videos to process from a single directory.
`check_video_paths`()	Validate and normalise video paths, returning a flat or nested list of mp4 paths.
`get_context_img_paths`(center_img_path)	Given the path to a center image frame, return paths of 5 context frames (n-2, n-1, n, n+1, n+2).
`split_video_files_by_view`(video_paths, ...)	For a list of videos from different sessions and views, split them up and return a list of lists like [[sess0_view0.mp4, sess0_view1.mp4, ...], [sess1_view0.mp4, sess1_view1.mp4, ...], ...]

PCA class to assist with computing PCA losses.

`convert_dict_values_to_tensors`(param_dict, ...)	Convert all values in a parameter dictionary to float tensors on the given device.
`format_multiview_data_for_pca`(data_arr, ...)	Reformat multiview data so each observation is a single body part across views.

`ComponentChooser`	Determine the number of PCA components to keep.
`EmpiricalEpsilon`	Find percentile value of a given loss tensor.
`KeypointPCA`	Class to collect data from a dataloader and compute PCA params.

Functions for predicting keypoints on labeled datasets and unlabeled videos.

`predict_dataset`(model, data_module, preds_file)	Save predicted keypoints for a labeled dataset.
`predict_video`()
`make_dlc_pandas_index`(cfg, keypoint_names)	Create a DLC-style three-level pandas MultiIndex for prediction DataFrames.
`generate_labeled_video`(video_file, preds_df, ...)	Overlay keypoint markers on a video and write the result to disk.

Convert batches of model outputs into a prediction dataframe.