Forward function for testing. Args: inputs (list[Tensor]): List of multi-level img features. img_metas (list[dict]): L

Forward function for testing. Args: inputs (list[Tensor]): List of multi-level img features. img_metas (list[dict]): L

Forward function for training mode. Args: x (list[Tensor]): Multi-level features from the upstream network, each

Forward function for training. Args: inputs (list[Tensor]): List of multi-level img features. img_metas (list[dict]):

Forward function for training. Args: inputs (list[Tensor]): List of multi-level img features. img_metas (list[dict]):

Returns: str: The latest checkpoint file in target directory.

Returns: The rank of the current process within the local (per-machine) process group.

Returns: The size of the per-machine process group, i.e. the number of processes per machine.

Generate grid points of multiple feature levels. Args: featmap_sizes (list[tuple]): List of feature map sizes in

Returns: bool: whether a checkpoint exists in the target directory.

Returns a dictionary of info about the object.

Returns a dictionary of info about the object.

Initialize the transformer weights.

Initialize weights.

ViT weight initialization, original timm impl (for reproducibility)

ViT weight initialization, original timm impl (for reproducibility)

Loss Name. This function must be implemented and will return the name of this loss function. This name will be used to combine differ

Loss function for outputs from a single decoder layer. Args: cls_scores (Tensor): Mask score logits from a single decoder layer

Calculate the CrossEntropy loss for masks. Args: pred (torch.Tensor): The prediction with shape (N, C), C is the number of cl

torch.Tensor: concatenated positive and negative boxes

list[int]: The number of priors (points) at a point on the feature grid

int: number of feature levels that the generator will be applied

Args: rng (None | int | numpy.random.RandomState): seed or state. kwargs (keyword arguments): - num_p

Resize pos_embed weights. Resize pos_embed using bicubic interpolate method. Args: pos_embed (torch.Tensor): Position emb

Directly returns the positive and negative indices of samples. Args: assign_result (:obj:`AssignResult`): Assigned results

Simple test with single image.

Simple test with single image.

Generate sparse points according to the ``prior_idxs``. Args: prior_idxs (Tensor): The index of corresponding anchors

Distributed synchronization of the metric Warning: does not synchronize the deque!

Returns the tokenized representation of given input string(s) Parameters ---------- texts : Union[str, List[str]]

The iteration step during training. This method defines an iteration step during training, except for the back propagation and optimi

The iteration step during training. This method defines an iteration step during training, except for the back propagation and optimi

The iteration step during validation. This method shares the same signature as :func:`train_step`, but used during val epochs. Note t

The iteration step during validation. This method shares the same signature as :func:`train_step`, but used during val epochs. Note t

Generate valid flags of points of multiple feature levels. Args: featmap_sizes (list(tuple)): List of feature map sizes in

Close to ViT-giant, with embed-dim 1536 and 24 heads => embed-dim per head 64

Args: x: (B, H, W, C) window_size (int): window size Returns: windows: (num_windows*B, window_size, window_size, C)

Args: windows: (num_windows*B, window_size, window_size, C) window_size (int): Window size H (int): Height of image

bool: whether the depther has auxiliary head

bool: whether the depther has decode head

bool: whether the depther has neck