Sample negative samples.

Sample positive samples.

Rezise the sample to ensure the given size. Keeps aspect ratio. Args: sample (dict): sample size (tuple): image size Returns

Computes one-to-one matching based on the weighted costs. This method assign each query prediction to a ground truth or background. T

Test with augmentations. Only rescale=True is supported.

Test with augmentations. Only rescale=True is supported.

torch.Tensor: concatenated positive and negative boxes

Calculate the binary CrossEntropy loss. Args: pred (torch.Tensor): The prediction with shape (N, 1). label (torch.Tensor): The le

Build backbone.

Builder of box coder.

Build depther.

Build head.

Build neck.

Build Transformer.

DINOv2 ViT-B/14 model (optionally) pretrained on the LVD-142M dataset.

Linear classifier (1 or 4 layers) on top of a DINOv2 ViT-B/14 backbone (optionally) pretrained on the LVD-142M dataset and trained on ImageNet-1k

DINOv2 ViT-B/14 model with registers (optionally) pretrained on the LVD-142M dataset.

Linear classifier (1 or 4 layers) on top of a DINOv2 ViT-B/14 backbone with registers (optionally) pretrained on the LVD-142M dataset and trained

DINOv2 ViT-g/14 model (optionally) pretrained on the LVD-142M dataset.

Linear classifier (1 or 4 layers) on top of a DINOv2 ViT-g/14 backbone (optionally) pretrained on the LVD-142M dataset and trained on ImageNet-1k

DINOv2 ViT-g/14 model with registers (optionally) pretrained on the LVD-142M dataset.

Linear classifier (1 or 4 layers) on top of a DINOv2 ViT-g/14 backbone with registers (optionally) pretrained on the LVD-142M dataset and trained

DINOv2 ViT-L/14 model (optionally) pretrained on the LVD-142M dataset.

Linear classifier (1 or 4 layers) on top of a DINOv2 ViT-L/14 backbone (optionally) pretrained on the LVD-142M dataset and trained on ImageNet-1k

Linear classifier (1 or 4 layers) on top of a DINOv2 ViT-L/14 backbone with registers (optionally) pretrained on the LVD-142M dataset and trained

DINOv2 ViT-S/14 model (optionally) pretrained on the LVD-142M dataset.

Linear classifier (1 or 4 layers) on top of a DINOv2 ViT-S/14 backbone (optionally) pretrained on the LVD-142M dataset and trained on ImageNet-1k

DINOv2 ViT-S/14 model with registers (optionally) pretrained on the LVD-142M dataset.

Linear classifier (1 or 4 layers) on top of a DINOv2 ViT-S/14 backbone with registers (optionally) pretrained on the LVD-142M dataset and trained

Enable distributed mode Args: set_cuda_current_device: If True, call torch.cuda.set_device() to set the current PyTorch CUDA

Placeholder for encode images with backbone and decode into a semantic depth map of the same size as input.

Extra repr.

Extra repr.

Placeholder for extract features from images.

Fix random seeds.

@x: (B,C,H,W)

@x: (B,C,H,W)

@x: (B,C,D,H,W)

@query: (B,L,C)

@cv: cost volume (B,C,D,H,W) @feat: (B,C,H,W)

@x: (B,N,D)

@cv: (B,C,D,H,W) where D is max disparity