Functions — github.com/UCSC-VLAA/OpenVision

hub / github.com/UCSC-VLAA/OpenVision / functions

Functions716 in github.com/UCSC-VLAA/OpenVision

⨍ Functions 716◇ Types & classes 90 ↳ Endpoints 1

Functionload_obrax_ckpt

(argv)

src/convert_upload/transfer_jax2hf.py:457

Methodlock

(self, unlocked_groups=0, freeze_bn_stats=False)

src/convert_upload/open_clip/modified_resnet.py:154

Methodlock

(self, unlocked_layers: int = 0, freeze_layer_norm: bool = True)

src/convert_upload/open_clip/hf_model.py:171

Methodlock

(self, unlocked_groups=0, freeze_bn_stats=False)

src/convert_upload/open_clip/transformer.py:543

Methodlock_image_tower

(self, unlocked_groups=0, freeze_bn_stats=False)

src/convert_upload/open_clip/model.py:258

Methodlock_image_tower

(self, unlocked_groups=0, freeze_bn_stats=False)

src/convert_upload/open_clip/model.py:346

Methodlock_text_tower

(self, unlocked_layers: int = 0, freeze_layer_norm: bool = True)

src/convert_upload/open_clip/model.py:350

Methodlog_timing_avg

Use this when you time sth multiple times per step (eg in a loop).

src/helpers/utils.py:424

Functionloss_fn

(params, images, labels)

src/main_clip.py:429

Functionloss_fn

(params_, imgs, lbls)

src/main_openvision2.py:277

Functionmae_loss

(*, pred, target, mask, norm_pix_loss: bool = True)

src/losses/common.py:327

src/main_openvision2.py:65

Functionmake

Returns gradient transform and learning rate functions.

src/optim/build_optax.py:188

Functionmake_for_inference

Makes an input pipeline for inference.

src/datasets/input_pipeline.py:141

Functionmake_mask_trees

Returns a boolean mask tree for every pattern (only first match).

src/helpers/utils.py:789

Methodmask_fn

(rng_i, x_i)

src/models/openvision2_model.py:127

Functionmatchfirst

(name, _)

src/helpers/utils.py:793

Functionmaybe_shard_and_put

(name, x)

src/datasets/input_pipeline.py:260

Functionmerge_params

Makes `loaded` pytree match `init`, warning or failing on mismatch. Args: loaded: pytree of parameters, typically loaded from a checkpoint.

src/models/common.py:543

Functionmixup

Perform mixup https://arxiv.org/abs/1710.09412. Args: rng: The random key to use.

src/transforms/mixup.py:190

Methodnum_channels

(self)

src/convert_upload/open_clip/transform.py:30

Methodnum_examples_per_process

(self, nprocess=None)

src/datasets/tfds.py:68

Functionpack_arg

Packs key-word args as a string to be parsed by `parse_arg()`.

src/configs/common.py:131

Functionpad_model_inputs

r"""Pad model input and generate corresponding input masks. `pad_model_inputs` performs the final packaging of a model's inputs commonly found in

src/transforms/bert_ops.py:191

Functionpad_shard_unpad

Wraps a function with code that pads, shards, then un-shards, un-pads. Args: wrapped: the function to be wrapped. Signature is `params, *args,

src/helpers/utils.py:55

Functionpad_shard_unpad_wrapper

(*args, min_device_batch=None, **kw)

src/helpers/utils.py:86

Functionparse

(x)

src/convert_upload/open_clip/utils.py:50

Functionparse_arg

Makes ConfigDict's get_config single-string argument more usable. Example use in the config file: import big_vision.configs.common as bvcc

src/configs/common.py:29

Methodper2x2_random_grid_indices

(mask)

src/models/vit.py:578

Methodper2x2_random_grid_mask

(mask, index)

src/models/vit.py:586

Functionpolicy_v0

Autoaugment policy that was used in AutoAugment Paper.

src/transforms/autoaugment.py:53

Functionpolicy_vtest

Autoaugment test policy for debugging.

src/transforms/autoaugment.py:88

Functionposemb_sincos_2d

Follows the MoCo v3 logic.

src/helpers/utils.py:911

Functionposterize

Equivalent of PIL Posterize.

src/transforms/autoaugment.py:233

Functionprefetch_scalar

(it, nprefetch=1, devices=None)

src/helpers/utils.py:203

Functionprepare_datasets

Returns unbatched `ds_images, ds_texts` datasets. Args: dataset: An image-text `tf.data.Dataset` that is expected to contain the followin

src/evaluators/proj/image_text/retrieval.py:75

Functionprepare_datasets

Returns unbatched `ds_images, ds_texts` datasets.

src/evaluators/proj/image_text/discriminative_classifier.py:69

Functionprofile

(name, ttl=3 * 365 * 24 * 3600, noop=False)

src/helpers/utils.py:810

Functionrag_tensor

(raw_tensor)

src/transforms/ops_general.py:219

Functionrandom_blur

Randomly blur an image. Args: image: `Tensor` representing an image of arbitrary size. height: Height of output image. width: Wi

src/transforms/simclr_aug.py:380

Functionrandom_brightness

Adjust the brightness of images by a random factor. Equivalent to `adjust_brightness()` using a `delta` randomly picked in the interval `[-max_de

src/transforms/ops_image.py:204

Functionrandom_color_jitter

(jitter_strength=0.5, impl='simclrv2')

src/transforms/ops_image.py:267

Functionrandom_color_jitter

(image, p=1.0, jitter_strength=0.5, impl='simclrv2')

src/transforms/simclr_aug.py:370

Functionrandom_crop_with_resize

Randomly crop and resize an image. Args: image: `Tensor` representing an image of arbitrary size. height: Height of output image.

src/transforms/simclr_aug.py:354

Functionrandom_mask_tokenize

( texts: Union[str, List[str]], context_length: int, sot_token_id: int, eot_to

src/convert_upload/open_clip/tokenizer.py:281

Methodrandom_masking

(self, x, mask_ratio, rng_mask=None)

src/models/text_transformer.py:709

Functionrecover_dtype

Numpy's `save` stores bfloat16 type as "void" type, so we recover it.

src/helpers/utils.py:681

Methodregister

Creates a function that registers its input.

src/helpers/registry.py:141

Functionregister_pooler

Decorator registering pooler class

src/convert_upload/open_clip/hf_model.py:39

Functionreplace_frozen

Replaces values matching frozen params in `pytree` with `replacement`.

src/optim/build_optax.py:178

Functionreplace_linear

(model, linear_replacement, include_modules=['c_fc', 'c_proj'], copy_weights=True)

src/convert_upload/open_clip/utils.py:65

Functionrotate

Rotates the image by degrees either clockwise or counterclockwise. Args: image: An image Tensor of type uint8. degrees: Float, a scalar angl

src/transforms/autoaugment.py:239

Methodrun

Computes all metrics.

src/evaluators/proj/image_text/contrastive.py:89

Methodrun

Returns metrics.

src/evaluators/proj/image_text/discriminative_classifier.py:461

Functionsaturation_foo

()

src/transforms/simclr_aug.py:153

Functionsave_checkpoint

Util for checkpointing: saves jax pytree objects to the disk. Args: checkpoint: arbitrary jax pytree to be saved. path: a path to save the

src/helpers/utils.py:693

Functionsave_checkpoint_ts

Preemption-safe saving of checkpoints using tssave.

src/helpers/utils.py:1024

Methodsave_pretrained

(self, dest)

src/convert_upload/open_clip/tokenizer.py:498

Methodsave_pretrained

(self, dest)

src/convert_upload/open_clip/tokenizer.py:575

Functionscan_ffn

(remat_ffn, carry, hidden_states)

src/models/bpt.py:27

Functionscan_kv_block

(carry, args)

src/models/bpt.py:106

Methodset_grad_checkpointing

(self, enable: bool = True)

src/convert_upload/open_clip/coca_model.py:144

Methodset_grad_checkpointing

(self, enable=True)

src/convert_upload/open_clip/modified_resnet.py:162

Methodset_grad_checkpointing

(self, enable=True)

src/convert_upload/open_clip/timm_model.py:143

Methodset_grad_checkpointing

(self, enable=True)

src/convert_upload/open_clip/hf_model.py:189

Methodset_grad_checkpointing

(self, enable=True)

src/convert_upload/open_clip/transformer.py:597

Methodset_grad_checkpointing

(self, enable=True)

src/convert_upload/open_clip/transformer.py:759

Methodset_grad_checkpointing

(self, enable=True)

src/convert_upload/open_clip/transformer.py:915

Methodset_grad_checkpointing

(self, enable=True)

src/convert_upload/open_clip/model.py:354

Implements Sharpness function from PIL using TF ops.

src/transforms/autoaugment.py:333

Functionshear_x

Equivalent of PIL Shearing in X dimension.

src/transforms/autoaugment.py:274

Functionshear_y

Equivalent of PIL Shearing in Y dimension.

src/transforms/autoaugment.py:285

Functionsigmoid_xent

(*, logits, labels, reduction=True)

src/losses/common.py:40

Functionsimple_mask_tokenize

( texts: Union[str, List[str]], context_length: int, sot_token_id: int, eot_to

src/convert_upload/open_clip/tokenizer.py:310

Functionskip_upper_half

(carry, args)

src/models/bpt.py:126

Functionsoftmax_xent

(*, logits, labels, mask=None, reduction=True, kl=False, axis=-1, smoothing=0.1)

src/losses/common.py:225

Functionsolarize

(image, threshold=128)

src/transforms/three_aug.py:172

Functionsolarize

(image, threshold=128)

src/transforms/autoaugment.py:187

Functionsolarize_add

(image, addition=0, threshold=128)

src/transforms/autoaugment.py:194

Methodsquare_mask

(start_height, start_width, mask_height, mask_width)

src/models/vit.py:558

Functionstart_input_pipeline

(data, config=None, mesh=None, data_sharding=None, mix_fn=None, tokenizer=None)

src/datasets/input_pipeline.py:242

Functionstart_input_pipeline_eval

(data, mesh=None, data_sharding=None, tokenizer=None, context_length=128)

src/datasets/input_pipeline.py:230

Functionstart_ragged_input_pipeline

(data, n_prefetch=1, shard=True, ragged=None)

src/datasets/input_pipeline.py:259

Functionstep_fn

Step to learning rate function.

src/optim/build_optax.py:114

Functionstrict_bool

(x)

src/configs/common.py:110

Functionsubstitute_prompt

(features)

src/evaluators/proj/image_text/discriminative_classifier.py:96

Functionsync

Syncs hosts and empties async computation queue.

src/helpers/utils.py:756

Functionsyntax_mask_tokenize

Returns the tokenized representation of given input string(s). Apply syntax masking before tokenize.

src/convert_upload/open_clip/tokenizer.py:332

Functiontemporary_ops

Registers specified pp ops for use in a `with` block. Example use: with pp_registry.remporary_ops( pow=lambda alpha: lambda d: {k: v**

src/helpers/registry.py:168

Methodtext_global_pool

(self, x, text: Optional[jnp.ndarray] = None, pool_type: str = 'argmax')

src/models/text_transformer.py:616

Functiontext_to_image_retrieval_eval

Runs the text-to-image retrieval eval from the distance matrix. Args: dist_matrix: Distance matrix between text and image embeddings (shape

src/evaluators/proj/image_text/image_text_retrieval.py:25

Functionto_grayscale

(image, keep_channels=True)

src/transforms/simclr_aug.py:48

Functiontokenize

(texts: Union[str, List[str]], context_length: int = DEFAULT_CONTEXT_LENGTH)

src/convert_upload/open_clip/tokenizer.py:277

Functiontokenizer_nltk

:param input_string: numpy.array, :return: tf.tensor string

src/transforms/bert_ops.py:395

Methodtotal_examples

Returns number of examples in the dataset, regardless of sharding.

src/datasets/core.py:44

Methodtotal_examples

(self)

src/datasets/tfds.py:64

Functiontrace_model

(model, batch_size=256, device=torch.device('cpu'))

src/convert_upload/open_clip/model.py:509

Functiontraining

Reads the data from a single dataset, or mixes it from multiple. The data is read either from one or mixed from multiple datasets, depending on t

src/datasets/input_pipeline.py:67

← previous next →601–700 of 716, ranked by callers