Subclass torch's LayerNorm to handle fp16.

Compose multiple transforms sequentially. Args: transforms (Sequence[dict | callable]): Sequence of transform object or confi

Collection of assign results.

Image to Patch Embedding.

Self-Attention Module. Args: in_channels (int): Input channels of key/query feature. channels (int): Output channels of key/query

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

ResLayer to build ResNet style backbone. Args: block (nn.Module): block used to build ResLayer. inplanes (int): inplanes of block

Atrous Spatial Pyramid Pooling (ASPP) Module. Args: dilations (tuple[int]): Dilation rate of each layer. in_channels (int): Input

Applies padding to input (if needed) so that input can get fully covered by filter you specified. It support two modes "same" and "corner". The

Applies padding adaptively to the input. This module can make input get fully covered by filter you specified. It support two modes "same" an

A wrapper of concatenated dataset. Same as :obj:`torch.utils.data.dataset.ConcatDataset`, but concat the group flag for image aspect ratio.

Load annotations for semantic segmentation. Args: reduce_zero_label (bool): Whether reduce all label value by 1. Usually used

Linear Embedding.

Standard points generator for multi-level (Mlvl) feature maps in 2D points-based detectors. Args: strides (list[int] | list[tuple[int

Pooling Pyramid Module used in PSPNet. Args: pool_scales (tuple[int]): Pooling scales used in Pooling Pyramid Module.

https://arxiv.org/abs/2212.00794

Convert some results to :obj:`torch.Tensor` by given keys. Args: keys (Sequence[str]): Keys that need to be converted to Tensor.

Applies padding to input (if needed) so that input can get fully covered by filter you specified. It support two modes "same" and "corner". The

Channel Attention Module (CAM)

Atrous Spatial Pyramid Pooling (ASPP) Module with depthwise separable conv.

A DistributedDataParallel wrapper for models in MMGeneration. In MMedting, there is a need to wrap different modules in the models with separ

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).

Vision Transformer with support for patch or hybrid CNN input stage

Implements feed-forward networks (FFNs) with identity connection. Args: embed_dims (int): The feature dimension. Same as `Mult

HuggingFace model adapter

HuggingFace tokenizer wrapper

A simple pipeline to load image.

Mask sampling result.

A ResNet class that is similar to torchvision's but contains the following changes: - There are now 3 "stem" convolutions as opposed to 1, wi

A wrapper for ``torch.nn.MultiheadAttention``. This module implements MultiheadAttention with identity connection, and positional encoding i

Normalize the image. Added key is "img_norm_cfg". Args: mean (sequence): Mean values of 3 channels. std (sequence): Std valu

Position Attention Module (PAM) Args: in_channels (int): Input channels of key/query feature. channels (int): Output channels of

Image to Patch Embedding. We use a conv layer to implement PatchEmbed. Args: in_channels (int): The num of input channels. Default:

Image to Patch Embedding

Radix Softmax module in ``SplitAttentionConv2d``. Args: radix (int): Radix of input. groups (int): Groups of input.

A wrapper of repeated dataset. The length of repeated dataset will be `times` larger than the original dataset. This is useful when the data

Resize images & seg. This transform resizes the input image to some scale. If the input dict contains the key "scale", then the scale in the

Squeeze-and-Excitation Module. Args: channels (int): The input (and output) channels of the SE layer. ratio (int): Squeeze ratio

Bbox sampling result. Example: >>> # xdoctest: +IGNORE_WANT >>> from mmdet.core.bbox.samplers.sampling_result import * # NOQA

Multi-headed attention from 'Attention Is All You Need' paper

r""" [`SiglipConfig`] is the configuration class to store the configuration of a [`SiglipModel`]. It is used to instantiate a Siglip model acc

r""" Constructs a SigLIP image processor. Args: do_resize (`bool`, *optional*, defaults to `True`): Whether to resize the

Multihead Attention Pooling.

r""" Constructs a Siglip processor which wraps a Siglip image processor and a Siglip tokenizer into a single processor. [`SiglipProcessor`] o

r""" This is the configuration class to store the configuration of a [`SiglipTextModel`]. It is used to instantiate a Siglip text encoder acco

Construct a Siglip tokenizer. Based on [SentencePiece](https://github.com/google/sentencepiece). This tokenizer inherits from [`PreTrainedTo

r""" This is the configuration class to store the configuration of a [`SiglipVisionModel`]. It is used to instantiate a Siglip vision encoder

Split-Attention Conv2d in ResNeSt. Args: in_channels (int): Same as nn.Conv2d. out_channels (int): Same as nn.Conv2d. ker

timm model adapter # FIXME this adapter is a work in progress, may change in ways that break weight compat

Implements one encoder layer in Vision Transformer. Args: embed_dims (int): The feature dimension. num_heads (int): Parallel atte