Gets default configs for CC12M dataset.

Gets default configs for CC12M dataset.

A wrapper function of different streaming function implementation. This is the place where we convert per-frame features (the input "features")

Get streaming features with intermadiate outputs. With N = self.num_dense_outputs, we currently forward the memory module N times, each time

DropPath Layer.

Get query embedding function. Args: module: OWL-ViT Flax module. variables: OWL-ViT variables. Returns: Jitted query embedding funct

Constructs training preprocess string.

Constructs training preprocess string.

Constructs training preprocess string.

Constructs training preprocess string.

Constructs training preprocess string.

Runs a single step of training. Note that in this code, the buffer of the second argument (batch) is donated to the computation. Args: fla

Runs a single step of training. Note that in this code, the buffer of the second argument (batch) is donated to the computation. Args: fla

Runs a single step of training. Note that in this code, the buffer of the second argument (batch) is donated to the computation. Args: fla

Generates a fake `batch`.

Returns config for testing MAE.

Provides a fixed position encoding for 2D and 3D coordinates. The embedding follows the initialisation method used in multiple papers such as "At

Returns GLUE task config.

Generate ground truth heatmaps and regression maps. Args: gt_boxes: array in shape (B x max_boxes x 4) gt_labels: array in shape (B x

Returns an image encoder.

Returns an image grid for summary writing.

Returns an image grid for summary writing.

Returns input shape for the dataset.

Returns temporal dims of input tokens for each view. Args: num_frames: Number of frames in the input video. view_configs: Configurations of

Obtain the instance mask from the blue channel of the segmentation file.

Return k nearest neighbours from dataset given queries. Args: queries: An [q, d] array where q is the number of queries, and d the dimens

Returns labels' weights to be used for computing weighted loss. This can used for weighting the loss terms based on the amount of available d

Returns labels' weights to be used for computing weighted loss. This can be used for weighting the loss terms based on the amount of availabl

Labels the inner products computed from `compute_inners` for verb loss.

Looks up for the learning rate scheduler and return lr_fn. Args: config: ConfigDict that has configuration ofthe learning rate function. Ret

Returns the LLaVA train source.

Returns the LLaVA train source.

Returns the localized narrative train source.

Returns the localized narrative train source.

Returns the localized narrative train source.

Returns the localized narrative train source.

Returns the localized narrative train source.

Returns the localized narrative train source.

Returns the localized narrative train source.

Returns the localized narrative train source.

Returns the localized narrative train source.

Returns the localized narrative train source.

Render boundary and wedge patches.

A convenience wrapper to all the loss_* functions in this class.

Arrange loss dictionary into a metrics dictionary.

Returns model class given its name.

Returns model class given its name.

Get the model class for the Universal Transformer project.

Returns model class given its name.

Get the model class for the PonderNet project.

Returns model class given its name.

Returns model class given its name.

Get the model class for the PlainViT project.

Selects Vivit model type.

Returns model class given its name.

Returns model class given its name.

Returns model class given its name.

Returns model class given its name.

Returns model class given its name.

Get the model class for the AdaTape project.

Returns model class given its name.

Returns model class given its name.

Make noise config strings to show metrics in XM.

Calculates the total number of training step and possibly steps_per_epoch. The main raining loop is based on number of training steps. Thus, for da

Obtain the number of examples in a thin DMVR or TFDS dataset.

Obtain the number of examples in a TFRecord dataset.

Obtain the number of examples in a thin DMVR or TFDS dataset.

Makes a Optax GradientTransformation for DeformableDETR.

Get optimizer for adversarial attack.

Calculates the per-patch importance function.

Calculates the per-pixel importance function.

Determines new labels after excluding bad classes. See Figure 1 in https://arxiv.org/abs/1604.01685 for which classes are excluded. Excluded clas

Get prediction function. Args: module: OWL-ViT Flax module. variables: OWL-ViT variables. Returns: Jitted predict function.