Estimates the Sim(3) transformation between `X` and `Y` point sets. Estimates c, R and t such as c * R @ X + t ~ Y. Parameters ----

Build the patch embed layer. If 'conv', we use a simple PatchEmbed conv layer. Otherwise, we use a vision transformer.

Open the ZIP file with mode read 'r', write 'w', exclusive create 'x', or append 'a'.

Initializes the position generator with an empty cache.

Build a pyramid of feature maps from the input. fmaps: Tensor (B, S, C, H, W) num_levels: number of pyramid levels (each dow

Return the rotation matrices for one of the rotations about an axis of which Euler angles describe, for each value of the angle given. A

Return the rotation matrices for one of the rotations about an axis of which Euler angles describe, for each value of the angle given. A

Computes frequency components for rotary embeddings. Args: dim: Feature dimension (must be even). seq_len: Maximum se

Return a tensor where each element has the absolute value taken from the, corresponding element of a, with sign taken from the corresponding

This function: - first downsizes the image with LANCZOS inteprolation, which is better than bilinear interpolation in

This function: - first downsizes the image with LANCZOS inteprolation, which is better than bilinear interpolation in

Run forward function and calculate loss for decode head in inference.

dynamic RoPE layers should recompute `inv_freq` in the following situations: 1 - growing beyond the cached sequence length (allow sca

dynamic RoPE layers should recompute `inv_freq` in the following situations: 1 - growing beyond the cached sequence length (allow sca

Get the number of images and videos for each sample to calculate the separation length of the sample tensor. These parameters are not

Check and initialize input transforms. The in_channels, in_index and input_transform must match. Specifically, when input_transform i

Parse the raw outputs (losses) of the network. Args: losses (dict): Raw output of the network, which usually contain

Process frame attention blocks. We keep tokens in shape (B*S, P, C).

Process global attention blocks. We keep tokens in shape (B, S*P, C).

Performs feature rotation by splitting and recombining feature dimensions. Args: x: Input tensor to rotate. Returns:

Returns torch.sqrt(torch.max(0, x)) but with a zero subgradient where x is 0.

Returns torch.sqrt(torch.max(0, x)) but with a zero subgradient where x is 0.

Transform inputs for decoder. Args: inputs (list[Tensor]): List of multi-level img features. Returns: Tensor

Process network output to extract 3D points and confidence values. Args: out: Network output tensor (B, C, H, W) activation:

Activate pose parameters with specified activation functions. Args: pred_pose_enc: Tensor containing encoded pose parameters [transl

### Parameters: - `points_src: torch.Tensor` of shape (..., N, 3) - `points_tgt: torch.Tensor` of shape (..., N, 3) - `weight: torch.

Applies Rotary Position Embedding to the query and key tensors, but only to the patch tokens, ignoring the prefix tokens (cls token and register t

Applies Rotary Position Embedding to the query and key tensors, but only to the patch tokens, ignoring the prefix tokens (cls token and register t

Converts a NumPy array to a PIL image. Automatically: - Squeezes dimensions of size 1. - Moves the channel dimension (if it has s

Associate two dictionaries of (stamp, data). As the time stamps never match exactly, we aim to find the closest match for every input tuple.

Placeholder for augmentation test.

Convert axis-angle representation (rotation vector) to rotation matrix, whose direction is the axis of rotation and length is the angle of rotation