Convenience class to aply the same operation to a whole set of images.

DPT Head for dense prediction tasks. This implementation follows the architecture described in "Vision Transformers for Dense Prediction"

Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).

2D image to patch embedding: (B,C,H,W) -> (B,N,D) Args: img_size: Image size. patch_size: Patch token size. in_chans

Residual convolution module.

Track a series of values and provide access to smoothed values over a window or the global series average.

CameraHead predicts camera parameters from token representations using iterative refinement. It applies a series of transformer blocks (the

Feature fusion block.

Artifically augmenting the size of a dataset.

Multi-task loss module that combines different loss types for VGGT. Supports: - Camera loss - Depth loss - Point loss -

Generates and caches 2D spatial positions for patches in a grid. This class efficiently manages the generation of spatial coordinates for patches

Artifically changing the size of a dataset.

2D Rotary Position Embedding implementation. This module applies rotary position embeddings to input tokens based on their 2D spatial positio

The Aggregator applies alternating-attention over input frames, as described in OniVGGT: Omni-Modality Driven Visual Geometry Grounded Transf

Computes and stores the average and current value. Args: name (str): Name of the metric being tracked device (torch.device, option

Define all basic options. Usage: class MyDataset (BaseStereoViewDataset): def _get_views(self, idx, rng): #

Random sampling under a constraint: each sample in the batch has the same feature, which is chosen randomly from a known pool of 'features' for

Concatenation of several datasets

a dataset that you can easily resize and combine. Examples: --------- 2 * dataset ==> duplicate each element 2x 10 @ dataset