Get the COCO keypoints and their left/right flip coorespondence map.

Get filename of latest checkpoint if it exists.

Process renderer part image to get body part indices.

Generate transformation matrix.

Process gt 3D keypoints and apply all augmentation transforms.

Compute 3D keypoint loss for the examples that 3D keypoint annotations are available. The loss is weighted by the confidence.

Compute 2D reprojection loss on the keypoints. The loss is weighted by the confidence. The available keypoints are different for each

Finds the bounding box parameters from the 2D keypoints. Args: kp (Kx3): 2D Keypoints. vis_thresh (float): Threshold for vis

Load a checkpoint.

Load a pretrained checkpoint. This is different from resuming training using --resume.

Process SMPL theta parameters and apply all augmentation transforms.

Read image, do preprocessing and possibly crop it according to the bounding box. If there are bounding box annotations, use them to crop the image

Convert quaternion coefficients to rotation matrix. Args: quat: size = [B, 4] 4 <===>(w, x, y, z) Returns: Rotation matrix cor

This function is borrowed from https://github.com/kornia/kornia Convert quaternion vector to angle axis of rotation. Adapted from ceres

Do Procrustes alignment and compute reconstruction error.

Dimension reduction by multi-layer perceptrons :param feature: list of [B, C_s, N] point-wise features before dimension reduction

Process rgb image and do augmentation.

Rotate axis angle parameters.

This function is borrowed from https://github.com/kornia/kornia Convert 3x4 rotation matrix to 4d quaternion vector This algorithm is b

Run evaluation on the datasets and metrics we report in the paper.

Run evaluation on the datasets and metrics we report in the paper.

https://github.com/soulslicer/STAF/tree/staf Don't forget to switch to 'staf' branch

Store all argument values to a json file. The default location is logs/expname/args.json.

Make axes of 3D plot have equal scale so that spheres appear as spheres, cubes as cubes, etc.. This is one possible solution to Matplotlib's

Compute per-vertex loss on the shape for the examples that SMPL annotations are available.

Applies median filtering and then gaussian filtering to bounding box parameters. Args: bbox_params (Nx3): [cx, cy, scale].

Rotate around x-axis by angle Args: category: 'torch' or 'numpy' prepend_dim: prepend an extra dimension Return: Rotation matrix with shape [1,

Rotate around y-axis by angle Args: category: 'torch' or 'numpy' prepend_dim: prepend an extra dimension Return: Rotation matrix with shape [1,

Rotate around z-axis by angle Args: category: 'torch' or 'numpy' prepend_dim: prepend an extra dimension Return: Rotation matrix with shape [1,

Render mesh using OpenDR verts: shape - (V, 3) faces: shape - (F, 3) img: shape - (224, 224, 3), range - [0, 255] (np.uint8)

Wrapper function for rendering process.

Retrieve dictionary entries