Resume from a checkpoint.

Load weights from .npz checkpoints for official Google Brain Flax implementation

Make causal mask used for bi-directional self-attention.

Adapted from Mesh Tensorflow: https://github.com/tensorflow/mesh/blob/0cb87fe07da627bf0b7e60475d59f95ed6b5be3d/mesh_tensorflow/transf

Load the best checkpoint for evaluation.

TODO: add other evaluation metrics for GQA

Use official VQA evaluation script to report metrics.

Save the checkpoint at the current epoch.

Saving the results in the format required for leaderboard evaluation.

Performs all_gather operation on the provided tensors. Graph remains connected for backward grad computation.

Compute 3D bounding box IoU. Input: corners1: numpy array (8,3), assume up direction is Z corners2: numpy array (8,3), assume up

Create by split datasets inheriting torch.utils.data.Datasets. # build() can be dataset-specific. Overwrite to customize.

Inverse function to angle2class. As ScanNet only has axis-alined boxes so angles are always 0.

Inverse function to size2class

Compute answer loss Args: data_dict: dict (read-only) Returns: answer_loss

Compute binned relative position bias

Compute 3D bounding box and semantic classification loss. Args: data_dict: dict (read-only) Returns: center_loss he

Computes the coordinates of the viewing frustum corresponding to one image and given camera parameters :param camera_to_world: torch

Computes the normal vectors (pointing inwards) to the 6 planes that bound the viewing frustum :param corner_coords: torch tensor of

Compute objectness loss for the proposals. Args: data_dict: dict (read-only) Returns: objectness_loss: scalar Tensor

Compute similarity i2t, t2i matrix for the given data loader.

Compute vote loss: Match predicted votes to GT votes. Args: data_dict: dict (read-only) Returns: vote_loss: scalar Tensor

calculate the confusion matrix.

Decay the learning rate

Apply elastic distortion on sparse coordinate space. pointcloud: numpy array of (number of points, at least 3 spatial dims) granu

Generic functions to compute precision/recall for object detection for multiple classes. Input: pred_all: map of {img_id:

Extract archive. Args: from_path: the path of the archive. to_path: the root path of the extracted files (directory of from_path)

pc: (N,3), box3d: (8,3)

extract CLIP text features.

FNV64-1A

Args: samples (dict): A dictionary containing the following keys: - image (torch.Tensor): A tensor of shape (batc