NOTE: this interface is experimental. Args: backbone: a backbone module, must follow detectron2's backbone interface

NOTE: this interface is experimental. Args: in_features (list[str]): list of names of input features to use

Args: size (int): the total number of data of the underlying dataset to sample from shuffle (bool): whether to shuffl

Args: image (ndarray): (H,W) or (H,W,C) ndarray of type uint8 in range [0, 255], or floating point in range [0, 1

Args: model, data_loader, optimizer: same as in :class:`SimpleTrainer`. grad_scaler: torch GradScaler to automaticall

The `__init__` method of any subclass should also contain these arguments. Args: in_channels (int): out_chan

Deformable convolution from :paper:`deformconv`. Arguments are similar to :class:`Conv2D`. Extra arguments: Args:

Args: cfg (CfgNode): num_gpus (int): if 0, will run on CPU

Args: cfg (CfgNode): num_gpus (int): if 0, will run on CPU

Add from_instances methods to the scripted Instances class.

Applies the exif orientation correctly. This code exists per the bug: https://github.com/python-pillow/Pillow/issues/3973 with the

Assert that each stride is 2x times its preceding stride, i.e. "contiguous in log2".

ONNX exported network doesn't have concept of device, assign necessary device option for each op in order to make it runable on GPU runtime.

compute the accuracy (soft score) of human answers

Configurations for some libraries.

Convert different format of boxes to an NxB array, where B = 4 or 5 is the box dimension.

Convert different format of masks or polygons to a tuple of masks and polygons. Returns: list[GenericMask]:

TODO (Xingyi): merge it with _create_heatmaps_from_dist dist: M x N return: heatmaps: M x 1

Creates gradient clipping closure to clip by value or by norm, according to the provided config.

dist: M x N labels: N return: heatmaps: M x C

Decode boxes and classification predictions of one featuer level, by the following steps: 1. filter the predictions based on

Args: masks: N, 1, H, W boxes: N, 4 img_h, img_w (int): skip_empty (bool): only paste masks within the region tha

Evaluate the box proposals in predictions. Fill self._results with the metrics for "box_proposals" task.

Evaluate the box proposals in predictions. Fill self._results with the metrics for "box_proposals" task.

Evaluate predictions. Fill self._results with the metrics of the tasks.

Evaluate predictions. Fill self._results with the metrics of the tasks. Args: predictions (list[dict]): list of outputs

Evaluate detection proposal recall metrics. This function is a much faster alternative to the official COCO API recall evaluation code. Howev

Evaluate detection proposal recall metrics. This function is a much faster alternative to the official LVIS API recall evaluation code. Howev

Evaluate the coco results using COCOEval API.

Args: iou_type (str): max_dets_per_image (None or int): limit on maximum detections per image in evaluating AP This l

Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.

Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.

Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.

Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token first unpad the input, th

Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token first unpad the input, th

Args: fields (dict[name: type])

Dynamically creates a new type that inherits the type of a given instance and overrides the `step` method to add gradient clipping

Statically infer the status of each blob, the status can be such as device type (CPU/GPU), layout (NCHW/NHWC), data type (float32/int8),

Get the arguments to be passed to ``aug.get_transform`` from the input ``aug_input``.