Stores assignments between predicted and truth boxes. Attributes: num_gts (int): the number of truth boxes considered when computing this

This class represents masks in the form of polygons. Polygons is a list of three levels. The first level of the list corresponds to objects,

This class represents masks in the form of bitmaps. Args: masks (ndarray): ndarray of masks in shape (N, H, W), where N is th

ResLayer to build ResNet style backbone for RPF in detectoRS. The difference between this module and base class is that we pass ``rfp_inplane

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

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

This class is almost the same as official pycocotools package. It implements some snake case function aliases. So that the COCO class has the

Cross Stage Partial Layer. Args: in_channels (int): The input channels of the CSP layer. out_channels (int): The output channels

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

ModulatedDeformConv2d with normalization layer used in DyHead. This module cannot be configured with `conv_cfg=dict(type='DCNv2')` because Dy

Feature Adaption Module. Feature Adaption Module is implemented based on DCN v1. It uses anchor shape prediction rather than feature map to

Wrapper for detector's inference with ONNXRuntime.

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

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

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

Bidirectional Corner Pooling Module (TopLeft, BottomRight, etc.) Args: in_channels (int): Input channels of module. out_channels

Detection block in YOLO neck. Let out_channels = n, the DetectionBlock contains: Six ConvLayers, 1 Conv2D Layer and 1 YoloLayer. The firs

Hourglass Module for HourglassNet backbone. Generate module recursively and use BasicBlock as the base unit. Args: depth (int): Dept

Data structure for instance-level annnotations or predictions. Subclass of :class:`GeneralData`. All value in `data_fields` should have the s

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

Shuffle the list and do it again after the list have traversed. The implementation logic is referred to https://github.com/wutong16/Distribut

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

Task decomposition module in task-aligned predictor of TOOD. Args: feat_channels (int): Number of feature channels in TOOD head.

Wrapper for detector's inference with TensorRT.

ASPP (Atrous Spatial Pyramid Pooling) This is an implementation of the ASPP module used in DetectoRS (https://arxiv.org/pdf/2006.02334.pdf)

An implementation of the absolute position embedding in PVT. Args: pos_shape (int): The shape of the absolute position embedding.

AdaptiveConv used to adapt the sampling location with the anchors. Args: in_channels (int): Number of channels in the input image

Basic residual block. This block is a little different from the block in the ResNet backbone. The kernel size of conv1 is 1 in this block whi

Bottleneck block for DilatedEncoder used in `YOLOF. <https://arxiv.org/abs/2103.09460>`. The Bottleneck contains three ConvLayers and one re

This wrapper is for loading the panoptic style annotation file. The format is shown in the CocoPanopticDataset class. Args: annotati

Center Weighting module to adjust the category-specific prior distributions. Args: force_topk (bool): When no point falls into gt_bbo

r"""Sampler that restricts data loading to the label of the dataset. A class-aware sampling strategy to effectively tackle the non-uniform cl

A wrapper of repeated dataset with repeat factor. Suitable for training on class imbalanced datasets like LVIS. Following the sampling strate

ConvUpsample performs 2x upsampling after Conv. There are several `ConvModule` layers. In the first few layers, upsampling will be applied af

The basic bottleneck block used in Darknet. Each ResBlock consists of two ConvModules and the input is added to the final output. Each ConvMo

Sampler that restricts data loading to a subset of the dataset. It is especially useful in conjunction with :class:`torch.nn.parallel.Distrib

DyHead Block with three types of attention. HSigmoid arguments in default act_cfg follow official code, not paper. https://github.com/microso

Dynamic ReLU (DyReLU) module. See `Dynamic ReLU <https://arxiv.org/abs/2003.10027>`_ for details. Current implementation is specialized for t

Focus width and height information into channel space. Args: in_channels (int): The input channels of this Module. out_channels (

High-Resolution Module for HRNet. In this module, every branch has 4 BasicBlocks/Bottlenecks. Fusion/Exchange is in this module.

Similar to `BatchSampler` warping a `DistributedSampler. It is designed iteration-based runners like `IterBasedRunner` and yields a mini-batch

Similar to `BatchSampler` warping a `GroupSampler. It is designed for iteration-based runners like `IterBasedRunner` and yields a mini-batch i

A fixed layer for calculating integral result from distribution. This layer calculates the target location by :math: `sum{P(y_i) * y_i}`, P(y

This is a module version of F.interpolate. Any arguments you give it just get passed along for the ride.

Inverted Residual Block. Args: in_channels (int): The input channels of this Module. out_channels (int): The output channels of t

Mask sampling result.

An implementation of MixFFN of PVT. The differences between MixFFN & FFN: 1. Use 1X1 Conv to replace Linear layer. 2. Introduce 3

A wrapper of multiple images mixed dataset. Suitable for training on multiple images mixed data augmentation like mosaic and mixup. For the a

Implements one encoder layer in PVT. Args: embed_dims (int): The feature dimension. num_heads (int): Parallel attention heads.

Merge patch feature map. This layer groups feature map by kernel_size, and applies norm and linear layers to the grouped feature map. Our imp

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

Random sampler. Args: num (int): Number of samples pos_fraction (float): Fraction of positive samples neg_pos_up (int, op

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

Res2Layer to build Res2Net style backbone. Args: block (nn.Module): block used to build ResLayer. inplanes (int): inplanes of blo

The basic residual block used in Darknet. Each ResBlock consists of two ConvModules and the input is added to the final output. Each ConvModule is

Display and save evaluation results. Args: show (bool): Whether to show the image. Default: True wait_time (float): Value of wait

Spatial pyramid pooling layer used in YOLOv3-SPP. Args: in_channels (int): The input channels of this Module. out_channels (int):

Shifted Window Multihead Self-Attention Module. Args: embed_dims (int): Number of input channels. num_heads (int): Number of atte

Smooth L1 loss. Args: beta (float, optional): The threshold in the piecewise function. Defaults to 1.0. reduction (st

An implementation of Spatial Reduction Attention of PVT. This module is modified from MultiheadAttention which is a module from mmcv.cnn.bric

Split-Attention Conv2d in ResNeSt. Args: in_channels (int): Number of channels in the input feature map. channels (int): Number o

Args: embed_dims (int): The feature dimension. num_heads (int): Parallel attention heads. feedforward_channels (int): The

Implements one stage in Swin Transformer. Args: embed_dims (int): The feature dimension. num_heads (int): Parallel attention head

Trident Convolution Module. Args: in_channels (int): Number of channels in input. out_channels (int): Number of channels in outpu

Window based multi-head self-attention (W-MSA) module with relative position bias. Args: embed_dims (int): Number of input channels.

YOLO anchor optimizer using differential evolution algorithm. Args: num_anchors (int) : Number of anchors. iters (int): Maximum i

r"""YOLO anchor optimizer using k-means. Code refer to `AlexeyAB/darknet. <https://github.com/AlexeyAB/darknet/blob/master/src/detector.c>`_.

Implementation of `ATSS <https://arxiv.org/abs/1912.02424>`_.

Assign a corresponding gt bbox or background to each bbox. Each proposals will be assigned with `0` or a positive integer indicating the grou

Bridging the Gap Between Anchor-based and Anchor-free Detection via Adaptive Training Sample Selection. ATSS head structure is similar with F

Handle empty batch dimension to AdaptiveAvgPool2d.

Albumentation augmentation. Adds custom transformations from Albumentations library. Please, visit `https://albumentations.readthedocs.io`

Anchor-free head (FCOS, Fovea, RepPoints, etc.). Args: num_classes (int): Number of categories excluding the background categ

Standard anchor generator for 2D anchor-based detectors. Args: strides (list[int] | list[tuple[int, int]]): Strides of anchors

Anchor-based head (RPN, RetinaNet, SSD, etc.). Args: num_classes (int): Number of categories excluding the background categor

Assign a corresponding gt bbox or background to each bbox. Each proposals will be assigned with an integer indicating the ground truth index

Associative Embedding Loss. More details can be found in `Associative Embedding <https://arxiv.org/abs/1611.05424>`_ and `CornerNet <http

Implementation of `AutoAssign: Differentiable Label Assignment for Dense Object Detection <https://arxiv.org/abs/2007.03496>`_.

AutoAssignHead head used in AutoAssign. More details can be found in the `paper <https://arxiv.org/abs/2007.03496>`_ . Args: for

Auto augmentation. This data augmentation is proposed in `Learning Data Augmentation Strategies for Object Detection <https://arxiv.org/pdf/1

Simplest RoI head, with only two fc layers for classification and regression respectively.

BBoxL1Cost. Args: weight (int | float, optional): loss_weight box_format (str, optional): 'xyxy' for DETR, 'xywh' for Sparse_R

Mixin class for testing det bboxes via DenseHead.

BFP (Balanced Feature Pyramids) BFP takes multi-level features as inputs and gather them into a single one, then refine the gathered feature

Balanced L1 Loss. arXiv: https://arxiv.org/pdf/1904.02701.pdf (CVPR 2019) Args: alpha (float): The denominator ``alpha`` in the bala

Base class for anchor optimizer. Args: dataset (obj:`Dataset`): Dataset object. input_shape (list[int]): Input image shape of the

Base assigner that assigns boxes to ground truth boxes.

Base bounding box coder.

Base class for DenseHeads.

Base class for detectors.

Exponential Moving Average Hook. Use Exponential Moving Average on all parameters of model in training process. All parameters have a ema bac