Args: img_size (int): Input image size. patch_size (int): Patch size. in_chans (int): Number of input ima

Return an activation function given a string

Return an activation function given a string

generate sine position embedding from a position tensor Args: pos_tensor (torch.Tensor): shape: [..., n]. num_pos_feats (int): pro

_summary_ Args: encoder_layer (_type_): _description_ num_layers (_type_): _description_ norm (_type_, op

Args: kernel_size (Tuple): kernel size of the projection layer. stride (Tuple): stride of the projection layer.

:param features: multi-scale features from the backbone :param masks: image mask :return: enhanced multi-scale features and m

Input: - memory: bs, \sum{hw}, d_model - memory_padding_mask: bs, \sum{hw} - spatial_shapes: nlevel, 2 Output:

Get relative positional embeddings according to the relative positions of query and key sizes. Args: q_size (int): size of qu

L2-normalize columns of X

Returns: dict[str->ShapeSpec]

Build a instance embedding branch from `cfg.MODEL.INS_EMBED_HEAD.NAME`.

Calculate absolute positional embeddings. If needed, resize embeddings and remove cls_token dimension for the original embeddings. Ar

:param reference: reference box coordinates from each decoder layer :param hs: content :param ref0: whether there are predict

:param reference: reference box coordinates from each decoder layer :param hs: content :param ref0: whether there are predict

Partition into non-overlapping windows with padding if needed. Args: x (tensor): input tokens with [B, H, W, C]. window_size

Window unpartition into original sequences and removing padding. Args: x (tensor): input tokens with [B * num_windows, window_size, w