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

↓ 1 callersFunctionbidirectional_contrastive_loss

Bidirectional contrastive losses (e.g. for contrastive trainer/evaluator).

src/losses/common.py:50

↓ 1 callersMethodbpe

(self, token)

src/convert_upload/open_clip/tokenizer.py:173

↓ 1 callersFunctionbuild_and_apply_nas_policy

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)`,

src/transforms/autoaugment.py:638

↓ 1 callersMethodbuild_attention_mask

(self)

src/convert_upload/open_clip/transformer.py:882

↓ 1 callersMethodbuild_causal_mask

(self)

src/convert_upload/open_clip/transformer.py:762

↓ 1 callersFunctionbuild_lut

(histo, step)

src/transforms/autoaugment.py:378

↓ 1 callersFunctioncenter_crop_or_pad

Center crops and/or pads the given image. If the image is torch Tensor, it is expected to have [..., H, W] shape, where ... means an arbitrary

src/convert_upload/open_clip/transform.py:167

↓ 1 callersFunctioncheck_and_compile_patterns

Validates and compiles a list of param-patterns. The validation consists of checking for common mistakes, currently only that the pattern does no

src/helpers/utils.py:763

↓ 1 callersFunctioncolor_jitter_nonrand

Distorts the color of the image (jittering order is fixed). Args: image: The input image tensor. brightness: A float, specifying the b

src/transforms/simclr_aug.py:78

↓ 1 callersFunctioncolor_jitter_rand

Distorts the color of the image (jittering order is random). Args: image: The input image tensor. brightness: A float, specifying the

src/transforms/simclr_aug.py:118

↓ 1 callersFunctionconvert_state_dict

(model: Union[CustomTextCLIP, CLIP], state_dict)

src/convert_upload/open_clip/convert.py:183

↓ 1 callersFunctionconvert_to_custom_text_state_dict

(state_dict: dict)

src/convert_upload/open_clip/model.py:432

↓ 1 callersFunctioncreate_learning_rate_schedule

Creates learning rate schedule, see (internal link). Args: total_steps: The total number of steps to run. batch_size: The global batch-size

src/optim/build_optax.py:83

↓ 1 callersFunctioncreate_schedule

(mult=1.0, **kw)

src/optim/build_optax.py:198

↓ 1 callersFunctioncrop_and_resize

Make a random crop and resize it to height `height` and width `width`. Args: image: Tensor representing the image. height: Desired ima

src/transforms/simclr_aug.py:294

↓ 1 callersFunctiondecode_variant

Converts a string like "B" or "B/32" into a params dict.

src/models/text_transformer.py:741

↓ 1 callersFunctiondecode_variant

Converts a model variant string into a dictionary of hyperparameters.

src/models/text_decoder_v2.py:480

↓ 1 callersFunctiondecode_variant

Converts a string like "B" or "B/32" into a params dict.

src/models/text_decoder.py:584

↓ 1 callersFunctiondefault_bpe

()

src/convert_upload/open_clip/tokenizer.py:28

↓ 1 callersFunctiondistorted_bounding_box_crop

Generates cropped_image using one of the bboxes randomly distorted. See `tf.image.sample_distorted_bounding_box` for more documentation. Args:

src/transforms/simclr_aug.py:244

↓ 1 callersFunctiondot_product_attention_weights

Computes dot-product attention weights given query and key. Used by :func:`dot_product_attention`, which is what you'll most likely use. But if y

src/models/common.py:53

↓ 1 callersMethodevaluate

Returns evaluation results.

src/evaluators/proj/image_text/retrieval.py:280

↓ 1 callersMethodevaluate

Returns evaluation results.

src/evaluators/proj/image_text/discriminative_classifier.py:389

↓ 1 callersFunctionevaluators

()

src/main_openvision2.py:354

↓ 1 callersFunctionfind_states

(opt_state, cls)

src/optim/build_optax.py:158

↓ 1 callersMethodflush_timings

(self)

src/helpers/utils.py:435

↓ 1 callersMethodforward

(self, image_features, text_features, logit_scale, output_dict=False)

src/convert_upload/open_clip/loss.py:120

↓ 1 callersFunctiongather_features

( image_features, text_features, local_loss=False, gather_with_grad=False,

src/convert_upload/open_clip/loss.py:19

↓ 1 callersFunctiongaussian_blur

Blurs the given image with separable convolution. Args: image: Tensor of shape [height, width, channels] and dtype float to blur. kern

src/transforms/simclr_aug.py:316

↓ 1 callersFunctiongaussian_filter2d

Perform Gaussian blur on image(s). Args: image: Either a 2-D `Tensor` of shape `[height, width]`, a 3-D `Tensor` of shape `[height,

src/transforms/three_aug.py:78

↓ 1 callersFunctiongenerate_readme

(model_card: dict, model_name: str)

src/convert_upload/open_clip/push_to_hf_hub.py:213

↓ 1 callersFunctionget_2d_sincos_pos_embed

grid_size: int of the grid height and width return: pos_embed: [grid_size*grid_size, embed_dim] or [1+grid_size*grid_size, embed_dim] (w/

src/convert_upload/open_clip/pos_embed.py:20

↓ 1 callersFunctionget_eval_fn

Produces eval function, also applies pmap.

src/evaluators/proj/image_text/contrastive.py:51

↓ 1 callersFunctionget_feature_value

(data, feature_name)

src/evaluators/proj/image_text/retrieval.py:102

↓ 1 callersMethodget_ground_truth

(self, device, dtype, num_logits, negative_only=False)

src/convert_upload/open_clip/loss.py:337

↓ 1 callersMethodget_logits

(self, image_features, text_features, logit_scale, logit_bias=None)

src/convert_upload/open_clip/loss.py:343

↓ 1 callersFunctionget_order

(x)

src/convert_upload/open_clip/tokenizer.py:350

↓ 1 callersMethodget_params

Get parameters

src/convert_upload/open_clip/transform.py:116

↓ 1 callersFunctionget_posemb

(self, typ, seqshape, width, name, dtype=jnp.float32, cls_token=False)

src/models/vit.py:166

↓ 1 callersFunctionget_posemb

( self, typ, max_len, width, name, dtype=jnp.float32,

src/models/text_transformer.py:64

↓ 1 callersFunctionget_reduction_mask_fn

Choose strategy for dropping (masking) tokens to achieve target context length

src/convert_upload/open_clip/tokenizer.py:391

↓ 1 callersMethodget_reference_weight

(self)

src/convert_upload/open_clip/transformer.py:306

↓ 1 callersFunctionget_type

Returns type of v and for boolean returns a strict bool function.

src/configs/common.py:107

↓ 1 callersFunctionimage_transform

( image_size: Union[int, Tuple[int, int]], is_train: bool, mean: Optional[Tuple[float,

src/convert_upload/open_clip/transform.py:274

↓ 1 callersMethodinit_parameters

(self)

src/convert_upload/open_clip/modified_resnet.py:141

↓ 1 callersMethodinit_parameters

(self)

src/convert_upload/open_clip/transformer.py:576

↓ 1 callersMethodinit_parameters

(self)

src/convert_upload/open_clip/transformer.py:735

↓ 1 callersFunctionjax_model_inference

(params, image, text)

src/convert_upload/transfer_jax2hf.py:555

↓ 1 callersFunctionlevel_to_arg

(hparams)

src/transforms/autoaugment.py:523

↓ 1 callersFunctionlevel_to_arg_increasing

(hparams)

src/transforms/autoaugment.py:551

↓ 1 callersFunctionlist_openai_models

Returns the names of available CLIP models

src/convert_upload/open_clip/openai.py:19

↓ 1 callersFunctionlist_pretrained_models_by_tag

return all models having the specified pretrain tag

src/convert_upload/open_clip/pretrained.py:526

↓ 1 callersFunctionload_any

we load checkpoint by given load_config

src/main_clip.py:231

↓ 1 callersFunctionload_big_vision_weights

Load weights from .npz checkpoints for official Google big_vision image-text models Currently the SigLIP source models are supported and a Custo

src/convert_upload/open_clip/convert.py:13

↓ 1 callersFunctionload_openai_model

Load a CLIP model Parameters ---------- name : str A model name listed by `clip.available_models()`, or the path to a model check

src/convert_upload/open_clip/openai.py:24

↓ 1 callersFunctionlocal_logits

(local_img, local_txt1, local_txt2)

src/losses/common.py:126

↓ 1 callersFunctionmake_final_policy

(tf_policy_)

src/transforms/autoaugment.py:671

↓ 1 callersFunctionneighbour_exchange

(from_rank, to_rank, tensor, group=None)

src/convert_upload/open_clip/loss.py:219

↓ 1 callersFunctionneighbour_exchange_bidir

(left_rank, right_rank, tensor_to_left, tensor_to_right, group=None)

src/convert_upload/open_clip/loss.py:239

↓ 1 callersFunctionneighbour_exchange_bidir_with_grad

(left_rank, right_rank, tensor_to_left, tensor_to_right, group=None)

src/convert_upload/open_clip/loss.py:303

↓ 1 callersFunctionnpload

Loads `fname` and returns an np.ndarray or dict thereof.

src/helpers/utils.py:126

↓ 1 callersMethodnum_examples_per_process

Returns a list of the numer of examples for each process. This is only needed for datasets that should go through make_for_inference. Args:

src/datasets/core.py:48

↓ 1 callersFunctionobrax_to_open_clip

(jax_params_cpu, H, W, Channel, pos_embed="learn", txt_model='vit', use_dense_general=True)

src/convert_upload/transfer_jax2hf.py:115

↓ 1 callersFunctiononehot

(labels, num_classes, on_value=1.0, off_value=0.0)

src/helpers/utils.py:119

↓ 1 callersMethodpad_and_add_class_token

(self, encoded_text, max_length)

src/convert_upload/open_clip/tokenizer.py:547

↓ 1 callersFunctionparse_name

Parses input to the registry's lookup function. Args: string_to_parse: can be either an arbitrary name or function call (optionally with

src/helpers/registry.py:71

↓ 1 callersFunctionposemb_sincos_1d

Follows the MoCo v3 logic.

src/models/text_transformer.py:45

↓ 1 callersFunctionposemb_sincos_2d

Follows the MoCo v3 logic.

src/convert_upload/transfer_jax2hf.py:98

↓ 1 callersFunctionprepare_inputs_for_generation

(input_ids, image_inputs, past=None, **kwargs)

src/convert_upload/open_clip/coca_model.py:479

↓ 1 callersFunctionpush_pretrained_to_hf_hub

( model_name, pretrained: str, repo_id: str, precision: str = 'fp32', image_mean: Optional

src/convert_upload/open_clip/push_to_hf_hub.py:160

↓ 1 callersFunctionpush_to_hf_hub

( model, tokenizer, model_config: Optional[dict], repo_id: str, commit_message: str = 'Add

src/convert_upload/open_clip/push_to_hf_hub.py:90

↓ 1 callersMethodrandom_masking

(self, x, mask_ratio, rng_mask=None, mask_mode='random', height=14, width=14)

src/models/vit.py:464

↓ 1 callersFunctionreference_attn

(query, key, value, causal, dtype)

src/models/bpt.py:201

↓ 1 callersFunctionrename

(k)

src/helpers/utils.py:653

↓ 1 callersFunctionresample_posemb

This function implements "high-res finetuning" for transformer models.

src/models/vit.py:838

↓ 1 callersFunctionresize_pos_embed

(state_dict, model, interpolation: str = 'bicubic', antialias: bool = True)

src/convert_upload/open_clip/model.py:525

↓ 1 callersFunctionresize_text_pos_embed

(state_dict, model, interpolation: str = 'linear', antialias: bool = False)

src/convert_upload/open_clip/model.py:559

↓ 1 callersFunctionsave_and_upload_for_hf

( model_state_dict, bv_conifg, text_model = 'vit' )

src/convert_upload/transfer_jax2hf.py:572

↓ 1 callersFunctionsave_config_for_hf

( model, config_path: str, model_config: Optional[dict] )

src/convert_upload/open_clip/push_to_hf_hub.py:40

↓ 1 callersFunctionsave_for_hf

( model, tokenizer: HFTokenizer, model_config: dict, save_directory: str, safe_serializati

src/convert_upload/open_clip/push_to_hf_hub.py:63

↓ 1 callersFunctionscale_values

(im)

src/transforms/autoaugment.py:314

↓ 1 callersFunctionscan_attention

(args)

src/models/bpt.py:102

↓ 1 callersFunctionselect_and_apply_random_policy

Select a random policy from `policies` and apply it to `image`.

src/transforms/autoaugment.py:625

↓ 1 callersMethodset_language

(self, src_lang)

src/convert_upload/open_clip/tokenizer.py:457

↓ 1 callersMethodset_language

(self, src_lang)

src/convert_upload/open_clip/tokenizer.py:596

↓ 1 callersFunctionset_model_preprocess_cfg

(model, preprocess_cfg: Dict[str, Any])

src/convert_upload/open_clip/model.py:608

↓ 1 callersFunctionshard_and_put

make local tf tensor to global jax array based on sharding strategy

src/datasets/input_pipeline.py:209

↓ 1 callersFunctionshould_remove

(k)

src/helpers/utils.py:660

↓ 1 callersMethodstem

(self, x)

src/convert_upload/open_clip/modified_resnet.py:166

↓ 1 callersFunctiontree_get

Get an entry of pytree by flattened key name, eg a/b/c, with nice error. Args: tree: the pytree to be queried. name: the path to extract fr

src/helpers/utils.py:593

↓ 1 callersFunctiontree_unflatten

Reverses `tree_flatten_with_names(tree)[0]`.

src/helpers/utils.py:538

↓ 1 callersFunctiontsload

Loads tensorstore-based array-tree from disk. If `tree` argument is provided, then array names to load and target structure is derived from the t

src/helpers/utils.py:1084

↓ 1 callersFunctiontssave

Save pytree using jax tensorstore-based checkpoint manager. NOTE: When overwriting an existing checkpoint with a different pytree, the result is,

src/helpers/utils.py:983

↓ 1 callersFunctionupdate_fn

Update step.

src/main_clip.py:404

↓ 1 callersFunctionupdate_fn

(state, batch, rng_key)

src/main_openvision2.py:265

FunctionModel

Factory function, because linen really don't like what I'm doing!

src/models/vit.py:811

FunctionModel

Factory function, because linen really don't like what I'm doing!

src/models/text_transformer.py:736

FunctionModel

Factory function to create the autoregressive model with a variant. Args: variant: A string shorthand for a model size (e.g., 'B', 'L').

src/models/text_decoder_v2.py:466

FunctionModel

Factory function, because linen really don't like what I'm doing!

src/models/text_decoder.py:579

Method__call__

Returns the tokenized representation of given input string(s) Parameters ---------- texts : Union[str, List[str]]

src/convert_upload/open_clip/tokenizer.py:227

← previous next →201–300 of 716, ranked by callers