SAM3PromptableConceptBackbone model

[source]

SAM3PromptableConceptBackbone class

keras_hub.models.SAM3PromptableConceptBackbone(
    vision_encoder,
    text_encoder,
    geometry_encoder,
    detr_encoder,
    detr_decoder,
    mask_decoder,
    dtype=None,
    **kwargs
)

A backbone for the Segment Anything Model 3 (SAM3).

SAM3 is a multi-modal model that supports text and geometry prompts (boxes) to perform object segmentation. It consists of a vision encoder, a text encoder, a geometry encoder for processing box prompts, and a DETR-based encoder-decoder architecture to fuse multi-modal features and predict segmentation masks.

Arguments

• vision_encoder: keras_hub.layers.SAM3VisionEncoder. A feature extractor for the input images.
• text_encoder: keras_hub.layers.SAM3TextEncoder. A Keras layer to compute embeddings for text prompts.
• geometry_encoder: keras_hub.layers.SAM3GeometryEncoder. A Keras layer to compute embeddings for geometry (box) prompts.
• detr_encoder: keras_hub.layers.SAM3DetrEncoder. A transformer-based encoder that fuses vision and prompt features.
• detr_decoder: keras_hub.layers.SAM3DetrDecoder. A transformer-based decoder that predicts object queries.
• mask_decoder: keras_hub.layers.SAM3MaskDecoder. A Keras layer to generate segmentation masks given the object queries and fused features.
• dtype: string or 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 in float32 precision regardless of dtype. Defaults to bfloat16.

Example

import numpy as np
import keras_hub

vision_encoder = keras_hub.layers.SAM3VisionEncoder(
    image_shape=(224, 224, 3),
    patch_size=14,
    num_layers=2,
    hidden_dim=32,
    intermediate_dim=128,
    num_heads=2,
    fpn_hidden_dim=32,
    fpn_scale_factors=[4.0, 2.0, 1.0, 0.5],
    pretrain_image_shape=(112, 112, 3),
    window_size=2,
    global_attn_indexes=[1, 2],
)
text_encoder = keras_hub.layers.SAM3TextEncoder(
    vocabulary_size=1024,
    embedding_dim=32,
    hidden_dim=32,
    num_layers=2,
    num_heads=2,
    intermediate_dim=128,
)
geometry_encoder = keras_hub.layers.SAM3GeometryEncoder(
    num_layers=3,
    hidden_dim=32,
    intermediate_dim=128,
    num_heads=2,
    roi_size=7,
)
detr_encoder = keras_hub.layers.SAM3DetrEncoder(
    num_layers=3,
    hidden_dim=32,
    intermediate_dim=128,
    num_heads=2,
)
detr_decoder = keras_hub.layers.SAM3DetrDecoder(
    image_shape=(224, 224, 3),
    patch_size=14,
    num_layers=2,
    hidden_dim=32,
    intermediate_dim=128,
    num_heads=2,
    num_queries=100,
)
mask_decoder = keras_hub.layers.SAM3MaskDecoder(
    num_upsampling_stages=3,
    hidden_dim=32,
    num_heads=2,
)
backbone = keras_hub.models.SAM3PromptableConceptBackbone(
    vision_encoder=vision_encoder,
    text_encoder=text_encoder,
    geometry_encoder=geometry_encoder,
    detr_encoder=detr_encoder,
    detr_decoder=detr_decoder,
    mask_decoder=mask_decoder,
)
input_data = {
    "pixel_values": np.ones((2, 224, 224, 3), dtype="float32"),
    "token_ids": np.ones((2, 32), dtype="int32"),
    "padding_mask": np.ones((2, 32), dtype="bool"),
    "boxes": np.zeros((2, 1, 5), dtype="float32"),
    "box_labels": np.zeros((2, 1), dtype="int32"),
}
outputs = backbone(input_data)

[source]

from_preset method

SAM3PromptableConceptBackbone.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:

1. a built-in preset identifier like 'bert_base_en'
2. a Kaggle Models handle like 'kaggle://user/bert/keras/bert_base_en'
3. a Hugging Face handle like 'hf://user/bert_base_en'
4. a ModelScope handle like 'modelscope://user/bert_base_en'
5. a path to a local preset directory like './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

• preset: string. A built-in preset identifier, a Kaggle Models handle, a Hugging Face handle, or a path to a local directory.
• load_weights: bool. If 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,
)