Add the cls token.

Add prediction of a single image to the evaluator. Args: prediction: Model prediction tuple of 4 arrays: boxes, scores, classes, ca

Adds functions to process integer feature to builders. This function expects the input to be either a `tf.train.SequenceExample` (with the featur

Add modality learned tokens.

Add positional encodings.

Compute box IoU. Boxes in format [l, t, w, h]. Args: boxes1: array in shape n x 4 boxes2: array in shape m x 4 Returns: iou: array in

Brevity penalty function for beam search penalizing short sequences. Args: alpha: float: brevity-penalty scaling parameter. length: int: le

Build a detection dataset.

Builds optimizer.

Loads a split from the COCO dataset using TensorFlow Datasets. Args: batch_size: int; The batch size returned by the data pipeline. train:

Loads a split from the COCO dataset using TensorFlow Datasets. Args: batch_size: int; The batch size returned by the data pipeline. train:

Loads a split from the COCO dataset using TensorFlow Datasets. Args: batch_size: The batch size returned by the data pipeline. train: Wheth

Merges residual connections.

Computes cost matrices for DeformableDETR predictions. Relevant code: https://github.com/fundamentalvision/Deformable-DETR/blob/11169a60c33333af0

Implements the matching cost matrix computations. Args: predictions: Dictionary of outputs from a model. Must contain 'pred_boxes'

Compute distance map using dictionary of shapes. Args: shapes: list of shape dictionaries, in order of back-object to front-object h: heigh

Compute the feature targets for feature regression. Args: batch: A single batch of data. This is updated with the feature target. config: T

Compute the feature targets for feature regression. Args: batch: A single batch of data. This is updated with the feature target. config: T

Loss and metrics function for single prediction layer.

Compute the maximum norm in a pytree of tensors.

Compute the maximum norm in a pytree of tensors.

Computes the metrics for all added predictions.

Compute HOTA on coco format. Args: gt_data: coco json format with key "annotations" and "images". pred_data: coco prediction format,

Evaluate metrics. Args: predictions: list of dict. Each dict is a prediction of an *instance*, with keys 'image_id', 'bbox', 'capti

Computes the metrics for all added predictions.

Computes precision.

Packs the COCOEval results into a dictionary. Args: coco_metrics: an array of length 12, as returned by `COCOeval.summarize()` Returns:

Generates split indices for each pyramid level.

Function creating FLT snippet encoding RPE. Computes the fourier transform (FT) in a give set of points. The FT is parameterized as a weighted su

Add more metadata to processors.crop_image.

Extract patches from an image. Args: im: Array of shape [N, H, W, C] patchsize: Tuple of integers representing the filter shape. stride

Convert yxyx [0, 1] normalized boxes to xyxy unnormalized format.

Decode a sequence of ASCII values into a string.

Converts segmentation to mask.

Computes the indicator functions from the distance functions. Args: dists: Array of shape [N, 2, R, R, H', W'] with samples of the two

Embedding input patches with 2D conv.

Extracts image features in the region of `desired_boxes_yxyx`. This works by taking the features of a bounding box with high IOU to the desi

Encode visual features. Args: features: (batch_size, num_tokens, dim) checkpoint_inds: (batch_size, num_caps_per_image) or None

Eval the model and write the summary.

Extract patches of size `rhs_shape` from `lhs`. Args: lhs: A 4-D Tensor; With shape `[batch, in_rows, in_cols, depth]. rhs_shape: tuple; S

Flattens the first two dimensions of a non-scalar array.

Flattens a dictionary, keeping empty leaves.

Fold patches into a single image. Args: patches: Array of shape [N, C, R, R, H', W'] val: Container of shape [N, C, H, W] Return

Forward second stage detection and get object features.

Forward point prediction.

Forward with points. Args: coords_input: (num_prompts, num_points, 2) image_size: (2,) Returns: embedding: (num_prompts, nu

Applies kernel attention with query, key, value tensors. This function defines the computation inside `call` with projected multi-head Q, K, V in

r"""Computes features based on an activation (e.g. ReLU-kernel by default). By default, computes random features for the ReLU kernel from http

Compute the best index for each patch. Has two possible modes determined by self.opts.parallel_mode: 1) When True, all N values are computed

Bottleneck representation. Args: x: input tensor. pooling_type: type of the classifier layer. Options are 'gap', 'gmp', 'gsp', 't

Returns a [num_classes, 3] array of colors for the model output labels.

Returns a list with the class names of the model output labels.

Returns a [num_classes] array of pixel frequency proportions.

Gets a closure to create a dataset.

Extract dataset name for eval_iter in xmanager measurements. Parent directory of the dataset files is used as its name. Args: dataset_path: P

Get streaming features with intermadiate outputs for all frames. Args: features: (video_batch_size, num_tot_tokens, dim) train: bool

Get drop pattern for drop layer.

Generate the drop-layer mask. Args: x: Input tensor. deterministic: Weather we are in the deterministic mode (e.g inference t

Return grid of 2D reference points within valid range by feature level. Args: spatial_shapes: [h, w] for each feature map level. valid_rati

Generates a fake `batch`.

Generates a fake `batch`. Returns: `batch`: Dictionary of None inputs and fake ground truth targets. outputs_noaux.pop('aux_outputs')

Generates a fake `batch`. Returns: `batch`: Dictionary of None inputs and fake ground truth targets. outputs_noaux.pop('aux_outputs')

Returns a callable metric function for global metrics. The return function implements metrics that require the prediction for the entire

Returns joint pairs of logits and labels. Args: logits: Tensor of shape [n, c] one_hot_targets: Tensor of shape [n, c] class_splits: Li

DropPath Layer.

DropPath Layer.

DropPath Layer.

DropPath Layer.

DropPath Layer.

Returns a {label: name} dict for a TFDS dataset.