Computes fewshot regression with eigenvalue solver in big_vision.

Set to zero all elements within dist of the outer boundary. Args: array: 2D array of non-zero elements dist: distance of boundary Return

Computes IoU between two sets of boxes. Boxes are in [x, y, x', y'] format [x, y] is top-left, [x', y'] is bottom right. Args: boxes1: Pre

Compute box IoU. Boxes in format [-l, -t, b, r]. Args: boxes1: array in shape B x n x 4 or n x 4 boxes2: array in shape B x m x 4 or m x 4

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

Predicts bounding boxes from image features. Args: image_features: Features extracted from the image, flattened into a 1d sequence

Sample points from boxes.

Build a policy from the given policies passed in and apply to image. Args: policies: list of lists of tuples in the form `(func, prob, level)`,

Build datasets.

Sequence to sequence flax module.

Generates a 2D grid of points evenly spaced in [0, 1] x [0, 1].

Generates a 2D grid of points evenly spaced in [0, 1] x [0, 1].

Returns CLIP's tokenizer.

Calculates and applies the rotary embedding.

Returns softmax cross entropy loss with an L2 penalty on the weights. Args: logits: Output of model in shape [batch, length, num_classes].

Initialize input embedding from a ViT model. This function is adapted from scenic.projects.vivit.google.model_utils. Here, we add support to inte

Run dp matching as https://github.com/fujiso/SODA/blob/master/soda.py.

Check all leaves are different in two PyTrees.

Compute Area Under the PR Curve for abnormal. Args: target: Numpy array of targets of shape (n_samples, n_classes). Since this metric is

Compute F1 score. Args: target: Numpy array of targets of shape (n_samples, n_classes). prediction: Numpy array of model predictions of sha

Compute precision. Args: target: Numpy array of targets of shape (n_samples, n_classes). prediction: Numpy array of model predictions of sh

Compute recall. Args: target: Numpy array of targets of shape (n_samples, n_classes). prediction: Numpy array of model predictions of shape

Compute Area Under the ROC Curve for abnormal. Args: target: Numpy array of targets of shape (n_samples, n_classes). Since this metric is

Compute recall. Args: target: Numpy array of targets of shape (n_samples, n_classes). prediction: Numpy array of model predictions of shape

Find the points where two circles intersect.

Find the points where the circle intersects with the line segment.

Creates a split from the Cityscapes dataset using TensorFlow Datasets. For the training set, we drop the last partial batch. This is fine to do b

Applies the class head to the image features. Args: image_features: Feature tokens extracted by the image embedder. query_embeddings:

Calculates metrics for the classification task. Currently we assume each metric_fn has the API: ```metric_fn(logits, targets, weights)``` an

Clears predictions/labels for previous run.

Clips boxes to image boundaries. It's called from roi_ops.py. Args: boxes: a tensor whose last dimension is 4 representing the coordinates

Convert custom tfrecord into tfds builder format.

Combines metrics across all sequences. Args: all_res: dict of dict; video_id to res from eval_sequence. Returns: res: same format

Computes average precision (detection task). Notes: The open source implementation is extremely slow, due to the way it iterates through the pr

Computes BERT loss. Args: mlm_logits: Masked LM logits of shape [batch, length, vocab_size]. nsp_logits: Next sentence prediction logits of s

Compute caption metrics.

Compute the COCO metrics for the collected annotations. Args: clear_annotations: if True, clears the `self.annotations` parameter a

Computes cost matrices for a batch of predictions. Relevant code: https://github.com/facebookresearch/detr/blob/647917626d5017e63c1217b99537deb2d

Computes the matching cost matrix. Args: predictions: Dictionary of outputs from a model. targets: Dictionary of ground truth targets

Compute few-shot metrics on one dataset.

Compute few-shot metrics on one dataset.

Compute few-shot metrics on one dataset.

Performs static analysis of the graph to compute theoretical FLOPs. This function is branched from scenic/common_lib/debug_utils.py. The differen

Computes a Fourier Transform in a given set of points. Computes the fourier transform (FT) in a give set of points. The FT is parameterized as a

Compute the intersections and vertices of a list of shapes.

Computes the IoU score between two temporal segments.

Computes the metrics for all added predictions.

Computes the relevant metrics for all added <target, output> pairs.

Computes the metrics for all added predictions.

Computes the metrics for all added predictions.

Computes the metrics for all added predictions.

Compute likelihood of a given tokenized sentence. This implements section 3.5 in the Dense VOC paper https://arxiv.org/pdf/2306.11729.pdf

Computes similarity scores and dump them directly instead of metrics.

Computes similarity scores and dump them directly instead of metrics.

Concatenate context tokens (e.g., input question) to visual tokens. Args: visual_features: (batch_size, feature_length, object_feat_size).

Concatenate context tokens (e.g., questions in QA) to visual tokens. Args: visual_features: (batch_size, feature_length, object_feat_size).

Concatenate context tokens (e.g., questions in QA) to visual tokens. Args: visual_features: (batch_size, feature_length, object_feat_size).

Creates a single tf.SequenceExample proto.

Creates a single tf.SequenceExample proto.

Create a new field reference which is controlled by the fn_flow. Args: fn_flow (fct): Function of signature (default) -> resolved_value which

Compute the shape of a conv given input shapes in canonical order. Args: lhs_shape: Input (lhs) shapes for which the output shapes should be in

Converts a big_vision checkpoint to a scenic train state. The model weights, global step and accumulated train time are extracted. Optimizer stat

Augment an image into a video.

Augment an image into a video.

Convert coco format to HOTA required format. Args: gt_data: coco json format with key "annotations" and "images". pred_data: coco pre

Convert COCO captioning format to detection by adding zero objects.

Adapt PyTorch model parameters to Jax ones. The steps are as follows. 1. PyTorch models have query-, key- and value-projections fused, whereas

Converts to COCO format.

Converts to COCO format.

Convert outputs to VidSTG evaluator's format.

Copy matched parameters from a restored one.

Copy matched parameters from a restored one.

Gives a scaling factor based on scheduling with a cosine decay. Args: step: int; Current step. steps_per_cycle: int; Number of steps to res