Normlize image by given mean and std.

Sparse tensor with support for both torchsparse and spconv backends. Parameters: - feats (torch.Tensor): Features of the sparse tens

Sparse tensor with support for both torchsparse and spconv backends. Parameters: - feats (torch.Tensor): Features of the sparse tens

MLP as used in Vision Transformer, MLP-Mixer and related networks

Resize sample to given size (width, height).

Prepare sample for usage as network input.

Embeds spatial positions into vector representations.

Feature fusion block.

An attention layer that allows for downscaling the size of the embedding after projection to queries, keys, and values.

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

Feature fusion block.

Embeds spatial positions into vector representations.

rgbs: (bsz, 3, h, w) torch tensor uint8 depths: (bsz, h, w) float32 bboxes: (bsz, 4) int K: (bsz, 3, 3) float32

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

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

Interpolation module.

A structure for storing masks and their related data in batched format. Implements basic filtering and concatenation.

Renderer for the Mesh representation. Args: rendering_options (dict): Rendering options. glctx (nvdiffrast.torch.RasterizeGL

MLP as used in Vision Transformer, MLP-Mixer and related networks

Transformer cross-attention block (MSA + MCA + FFN) with adaptive layer norm conditioning.

Embeds scalar timesteps into vector representations.

The Aggregator applies alternating-attention over input frames, as described in VGGT: Visual Geometry Grounded Transformer. Args:

Modification of https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/levit.py: ConvNorm such that ConvTranspose2

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

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

Renderer for the Voxel representation. Args: rendering_options (dict): Rendering options.

Renderer for the Voxel representation. Args: rendering_options (dict): Rendering options.

Network for monocular depth estimation.

Network for monocular depth estimation.

MLP as used in Vision Transformer, MLP-Mixer and related networks

MLP as used in Vision Transformer, MLP-Mixer and related networks

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

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