CLIPBackbone
classkeras_hub.models.CLIPBackbone(
vision_encoder, text_encoder, projection_dim, dtype=None, name=None, **kwargs
)
CLIP core network with hyperparameters.
This backbone implements the base architecture for Contrastive Language-Image Pretraining (CLIP) model. It includes a vision and text encoders and the corresponding projection layers. This backbone will output the final logit scores corresponding to each image and token input. These values are cosine similarities between the corresponding image and text features.
The default constructor gives a fully customizable, randomly initialized
CLIP model with any number of layers, heads, and embedding dimensions. To
load preset architectures and weights, use the from_preset
constructor.
Arguments
keras.mixed_precision.DTypePolicy
. The dtype to use
for the models computations and weights. Note that some
computations, such as softmax and layer normalization will always
be done a float32 precision regardless of dtype.Example
input_data = {
"images": np.ones(shape=(1, 224, 224, 3), dtype="float32"),
"token_ids": np.ones(shape=(1, 12), dtype="int32"),
}
# Pretrained CLIP model.
model = keras_hub.models.CLIPBackbone.from_preset("clip_vit_base_patch32")
model(input_data)
# Randomly initialized CLIP model with custom config.
vision_encoder = keras_hub.models.CLIPVisionEncoder(
patch_size=32,
hidden_dim=768,
num_layers=8,
num_heads=8,
intermediate_dim=2048,
image_shape=(384, 384, 3),
)
text_encoder = keras_hub.models.CLIPTextEncoder(
vocabulary_size=49408,
embedding_dim=768,
hidden_dim=768,
num_layers=8,
num_heads=8,
intermediate_dim=2048,
)
model = keras_hub.models.CLIPBackbone(
vision_encoder=vision_encoder,
text_encoder=text_encoder,
projection_dim=256,
)
model(input_data)
from_preset
methodCLIPBackbone.from_preset(preset, load_weights=True, **kwargs)
Instantiate a keras_hub.models.Backbone
from a model preset.
A preset is a directory of configs, weights and other file assets used
to save and load a pre-trained model. The preset
can be passed as a
one of:
'bert_base_en'
'kaggle://user/bert/keras/bert_base_en'
'hf://user/bert_base_en'
'./bert_base_en'
This constructor can be called in one of two ways. Either from the base
class like keras_hub.models.Backbone.from_preset()
, or from
a model class like keras_hub.models.GemmaBackbone.from_preset()
.
If calling from the base class, the subclass of the returning object
will be inferred from the config in the preset directory.
For any Backbone
subclass, you can run cls.presets.keys()
to list
all built-in presets available on the class.
Arguments
True
, the weights will be loaded into the
model architecture. If False
, the weights will be randomly
initialized.Examples
# Load a Gemma backbone with pre-trained weights.
model = keras_hub.models.Backbone.from_preset(
"gemma_2b_en",
)
# Load a Bert backbone with a pre-trained config and random weights.
model = keras_hub.models.Backbone.from_preset(
"bert_base_en",
load_weights=False,
)
Preset | Parameters | Description |
---|---|---|
clip_vit_base_patch16 | 149.62M | 150 million parameter, 12-layer for vision and 12-layer for text, patch size of 16, CLIP model. |
clip_vit_base_patch32 | 151.28M | 151 million parameter, 12-layer for vision and 12-layer for text, patch size of 32, CLIP model. |
clip_vit_b_32_laion2b_s34b_b79k | 151.28M | 151 million parameter, 12-layer for vision and 12-layer for text, patch size of 32, Open CLIP model. |
clip_vit_large_patch14 | 427.62M | 428 million parameter, 24-layer for vision and 12-layer for text, patch size of 14, CLIP model. |
clip_vit_large_patch14_336 | 427.94M | 428 million parameter, 24-layer for vision and 12-layer for text, patch size of 14, image size of 336, CLIP model. |
clip_vit_h_14_laion2b_s32b_b79k | 986.11M | 986 million parameter, 32-layer for vision and 24-layer for text, patch size of 14, Open CLIP model. |
clip_vit_g_14_laion2b_s12b_b42k | 1.37B | 1.4 billion parameter, 40-layer for vision and 24-layer for text, patch size of 14, Open CLIP model. |
clip_vit_bigg_14_laion2b_39b_b160k | 2.54B | 2.5 billion parameter, 48-layer for vision and 32-layer for text, patch size of 14, Open CLIP model. |