Assign gt to bboxes. The process is very much like the max iou assigner, except that positive samples are constrained within the cell

switch pipeline.

Calculate the accuracy. Args: pred (torch.Tensor | np.ndarray | Sequence): The prediction results. It can be labe

Construct a sequential conv, bn and relu. Args: in_channels (int): Dimension of input features. out_channels (int): Dimension of

Coerces input into a random number generator. If the input is None, then a global random state is returned. If the input is a numeric value,

Evaluate the model performance of the whole dataset after processing all batches. Args: size (int): Length of the entire

`Focal Loss <https://arxiv.org/abs/1708.02002>`_ for targets in gaussian distribution. Args: pred (torch.Tensor): The prediction.

Get annotation by index. Args: idx (int): Global index of ``KFoldDataset``. Returns: dict: The idx-th annota

Convert a label to onehot format tensor. Args: label (LABEL_TYPE): Label value. num_classes (int): The number of classes. Re

Make divisible function. This function rounds the channel number down to the nearest value that can be divisible by the divisor. Args:

Calculate the dist matrixes for loss measurement. Args: key_embeds (Tensor): Embeds of positive bboxes in sampling results

Merge augmented mask prediction. Args: aug_masks (list[Tensor]): each has shape (n, c, h, w). img_metas (dict): Image

NMS for multi-class bboxes. Args: multi_bboxes (Tensor): shape (n, #class*4) or (n, 4) multi_scores (Tensor): shape (n, #class),

nn.Module: normalization layer after the third convolution layer

Scale boxes with type of tensor or box type. Args: boxes (Tensor or :obj:`BaseBoxes`): boxes need to be scaled. Its type can

Set user-defined new property.

Transpose the dimensions of `x`.

Apply element-wise weight and reduce loss. Args: loss (Tensor): Element-wise loss. weight (Tensor): Element-wise weights.

Get the pad_shape of each image based on data and pad_size_divisor.

Handle empty batch dimension to adaptive_avg_pool2d. Args: input (tensor): 4D tensor. output_size (int, tuple[int,int]): the targ

Regularly check whether the loss is valid every n iterations. Args: runner (:obj:`Runner`): The runner of the training process.

Pad each input to `max_len`, then stack them. If `max_len` is None, then it is the max size of the first dimension of each input. https:/

Get offset for deformable conv based on anchor shape NOTE: currently support deformable kernel_size=3 and dilation=1 Args:

Computes 1-to-k gt-pred matching based on the weighted costs. This method assign each query prediction to a ground truth or backgroun

Assign gt to bboxes. This method assigns gts to every prior (proposal/anchor), each prior will be assigned with -1, or a semi-positiv

Check if norm layer is in correct train state.

Construct a sequential conv and bn. Args: in_channels (int): Dimension of input features. out_channels (int): Dimension of output

Initialize weights.

Perform forward propagation of the head, then calculate loss and predictions from the features and data samples. Over-write because im

Calculate the loss based on the features extracted by the detection head. Args: cls_scores (list[Tensor]): Box scores for

make plugins for block. Args: in_channels (int): Input channels of plugin. plugins (list[dict]): List of plugins cfg

Process one batch of data samples and predictions. The processed results should be stored in ``self.results``, which will be used to c

Refine bboxes during training. Args: sampling_results (List[:obj:`SamplingResult`] or List[:obj:`InstanceData`]):

Remove the target transform type from the pipeline. Args: pipeline (List[dict] | List[BaseTransform]): The transforms list. targe

Rescale boxes w.r.t. rescale_factor in-place. Note: Both ``rescale_`` and ``resize_`` will enlarge or shrink boxes w.

Get current active map with (B, 1, f, f) shape or index format.

Get adaptive scales according to areas. The scale range is [0.5, 1.0]. When the area is less than ``min_area``, the scale is 0.5 while the ar

Get decoder layers according to attribute name.

Initialize bbox regression conv layers of the head.

Draw datasample and save to all backends. - If GT and prediction are plotted at the same time, they are displayed in a stitched image

Assign gt to anchors. This method assign a gt bbox to every bbox (proposal/anchor), each bbox will be assigned with -1, 0, or a posit

Undo batch shuffle. Args: x (torch.Tensor): Data in each GPU. idx_unshuffle (torch.Tensor): Index for restoring. Returns:

Close mosaic and mixup augmentation and switches to use L1 loss.

Reload ``device`` from self.tensor.

Calculate feature similarity from embeddings. Args: key_embeds (Tensor): Shape (N1, C). ref_embeds (Tensor): Shape (N2, C).

Extract features. Args: batch_inputs (Tensor): Image tensor, has shape (bs, dim, H, W). Returns: tuple[Tenso

Generate the coordinate. Args: featmap_sizes (tuple): The feature to be calculated, of shape (N, C, W, H). device (st

filtering parameters. default skips that filter. :param catNms (str array) : get cats for given cat names :param supNms (str array)

Get adaptive scale according to image shape. The target scale depends on the the short edge length of the image. If the short edge length equ

Get COCO category ids by index. Args: idx (int): Index of data. Returns: List[int]: All categories in the im

Get the expansion of a residual block. The block expansion will be obtained by the following order: 1. If ``expansion`` is given, just retur

Get user-defined property.