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

Linear head for dust3r Each token outputs: - 16x16 3D points (+ confidence)

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

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

Linear head for dust3r Each token outputs: - 16x16 3D points (+ confidence)

Linear Embedding

Image to Patch Embedding

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

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

Construct the CLS token, mask token, register tokens, position and patch embeddings.

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

return positions of patches

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).

Residual convolution module.

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

A conv block that bundles conv/norm/activation layers. This block simplifies the usage of convolution layers, which are commonly used with a

Wrapped DINOv2 encoder supporting gradient checkpointing. Input is RGB image in range [0, 1].

Multi-headed attention compatible with ALL_ATTENTION_FUNCTIONS.

Multi-headed attention compatible with ALL_ATTENTION_FUNCTIONS.

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).

Construct the CLS token, mask token, position and patch embeddings.

Construct the CLS token, mask token, position and patch embeddings.

This corresponds to the Block class in the original implementation.

This corresponds to the Block class in the original implementation.

r""" This is the configuration class to store the configuration of a [`Dinov2WithRegistersModel`]. It is used to instantiate an Dinov2WithRegi