Render a skeleton, defined by its xyz coordinates in a nx3 numpt array and corresponding properties in skeleton_cls :param skeleton: A nx3 n

Scale a given heatmap with the given strategy and temperature :param heatmap: Tensor of any dimensionality to be scaled :param strategy:

Calculate the mpjpe (l2) between 3D prediction and target vectors with numpy :param prediction: Predicted samples, as num_samples x 3 :p

Refine a predicted skeleton, based on its input skeleton, the heatmap, and the given weights :param predicted_skeleton: The predicted skeleto

Validation, once every epoch :param model: The model being trained :param data: The dataloader, in val mode :param heatmap_criterion

Calculate the diversity between all k sampled predictions :param path: Path containing outputs from sample.py: results/ and, if refined, ref

Easy-to-use HSV to RGB color space transformation :param h: Hue :param s: Saturation :param v: Value :return: The RGB equivalent

Start training. The configuration can be found in config.yaml and will be parsed by hydra :param conf: The configuration, populated by hydra

Start sampling given experiment (either grab or h36m) with the given base_path :param experiment: The experiment to sample: grab or h36m, ha

Run refinement of which will improve the quality of predicted skeletons by enforcing constraints on bone lengths, joint angles, end effector

Start evaluation: Qualitatively by rendering skeleton predictions; quantitatively by calculating mpjpe and diversity :param experiment: The e

Generate the characteristic 3d pose data from the Human3.6M dataset Instruction on how to download and extract Human3.6M data are in README.m

Generate a numpy npz file containing the xyz joint locations of all characteristic 3d poses in our dataset :param input_path: The path conta

Calculate the top-1 mpjpe :param path: Path containing outputs from sample.py: results/ and, if refined, refined/ :param sample_ids: Lis

Render input, predicted, and target skeletons using pyrender :param path: Path containing outputs from sample.py: results/ and, if refined,

Find the rotation matrix that aligns vec1 to vec2 :param vec1: A 3d "source" vector :param vec2: A 3d "destination" vector :return A

Sample from predicted heatmaps, sample by sample :param models: A dict of models for right hand, left hand, and rest of the body; checkpoint

Generates a string, summarizing a pytorch model for an easy overview over sub-modules and parameter count :param model: The pytorch model (s

Training, once every epoch :param model: The model being trained :param data: The dataloader, in train mode :param heatmap_criterion

Initialize the PoseHeatMapCriterion :param conf: The full training configuration :param small: Whether this criterion is use

Initialize the HeatmapOffsetCriterion :param conf: The full training configuration

Initializes the JointSampler: Construct box filter, set parameters :param heatmap_resolution: The spatial resolution of the heatmap,

Build the model for characteristic 3d pose prediction, consisting of an encoder, attention module, and volumetric decoder :param con

Characteristic Pose dataset, built on top of Human3.6M :param phase: The experiment phase: train, val, test. Affects: sample ids, au

Characteristic Pose dataset, built on top of GRAB :param phase: The experiment phase: train, val, test. Affects: sample ids, augment

Sample from given heatmap and add offsets (No batching, only single heatmap). Sample k samples. Volume will be centered at volume_center or a

Creates a 1D vector filled with zeros with given size and a one at given idx :param idx: Where to place the one :param size: 1D size of

Creates a 1D vector filled with zeros with given size and a one at given idx :param idx: Where to place the one :param size: 1D size of