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

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

Levenberg-Marquardt solver, with fixed number of iterations. - For 4DoF case, the pose is parameterized as [x, y, z, yaw], where yaw is the

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

Random Sample Levenberg-Marquardt solver, a generalization of RANSAC. Used for initialization in ambiguous problems.

Interpolation module.

DPT module for CroCo. by default, hooks_idx will be equal to: * for encoder-only: 4 equally spread layers * for encoder+decoder: last enc

Residual convolution module.

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

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

Dataset of outdoor street scenes, 5 images each time

Concatenation of several datasets

DPT output adapter. :param num_cahnnels: Number of output channels :param stride_level: tride level compared to the full-sized image.

Adapt croco's DPTOutputAdapter implementation for dust3r: remove duplicated weigths, and fix forward for dust3r

End-to-End Probabilistic Perspective-n-Points for 6DoF pose estimation. The pose is parameterized as [x, y, z, w, i, j, k], where [w, i, j, k

Feature fusion block.

L1 (Manhattan) distance between points

Euclidean distance between 3d points

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

Log L1 distance between points: first apply log transform, then L1 distance Helps with scale variations in the input space

Handle images with non-square aspect ratio. All images in the same batch have the same aspect ratio. true_shape = [(height, width) ..

Mixed L1 and LogL1 loss: L1 for x,y coordinates and LogL1 for z coordinate

Artifically augmenting the size of a dataset.

This a Dummy Optimizer. To use only when the goal is to visualize the results for a pair of images (with is_symmetrized)

just adding _init_weights + position getter compared to timm.models.layers.patch_embed.PatchEmbed

DPT module for dust3r, can return 3D points + confidence for all pixels

return positions of patches

return positions of patches

random masking

Artifically changing the size of a dataset.

Smooth L1 loss between points

Dataset of indoor scenes, 5 images each time

Dataset of outdoor street scenes, 5 images each time

Two siamese encoders, followed by two decoders. The goal is to output 3d points directly, both images in view1's frame (hence the asymmetry).

Two siamese encoders, followed by two decoders. The goal is to output 3d points directly, both images in view1's frame (hence the asymmetry).

Optimize a global scene, given a list of pairwise observations. Graph node: images Graph edges: observations = (pred1, pred2)

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

Weighted regression by learned confidence. Assuming the input pixel_loss is a pixel-level regression loss. Principle: high-confi

A dataclass used to load annotations from json.

End-to-End Probabilistic Perspective-n-Points. Args: mc_samples (int): Number of total Monte Carlo samples num_iter (int): N

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