Description #

ViT image semantic embedding processor, suitable for image similarity search, content retrieval, and related scenarios.

Key features #

• Multi-model support:
  • Google official ViT models
  • Meta DINOv2 vision models
• Feature extraction modes:
  • CLS token embedding vector
  • Global average pooling
• Input format compatibility:
  • URL
  • Base64 encoding
  • Binary stream
• Performance optimization:
  • FP16 inference acceleration
  • Multi-GPU parallel computation

Fault tolerance mechanism #

• Returns an all-zero vector when extraction fails

Recommended models #

• google/vit-base-patch16-224-in21k (768 dimensions)
• google/vit-large-patch16-224-in21k (1024 dimensions)
• facebook/dinov2-base (768 dimensions)
• facebook/dinov2-large (1024 dimensions)

Operator parameters #

Input #

Output #

An array containing feature vectors. Each element is a nested array of float type,
The array dimensions are determined by the model output

Parameters #

If a parameter does not have a default value, it is required

Examples #

The following code demonstrates how to use daft to run the operator and compute image embeddings.

from __future__ import annotations

import logging
import os

import ray

import daft
from daft import col
from daft.las.functions.image.embedding.image_vit_embedding import ImageViTEmbedding
from daft.las.functions.udf import las_udf

if __name__ == "__main__":

    if os.getenv("DAFT_RUNNER", "ray") == "ray":

        def configure_logging():
            logging.basicConfig(
                level=logging.INFO,
                format="%(asctime)s - %(name)s - %(levelname)s - %(message)s",
                datefmt="%Y-%m-%d %H:%M:%S.%s".format(),
            )
            logging.getLogger("tracing.span").setLevel(logging.WARNING)
            logging.getLogger("daft_io.stats").setLevel(logging.WARNING)
            logging.getLogger("DaftStatisticsManager").setLevel(logging.WARNING)
            logging.getLogger("DaftFlotillaScheduler").setLevel(logging.WARNING)
            logging.getLogger("DaftFlotillaDispatcher").setLevel(logging.WARNING)

        import ray

        ray.init(dashboard_host="0.0.0.0", runtime_env={"worker_process_setup_hook": configure_logging})
        daft.set_runner_ray()

    daft.set_execution_config(actor_udf_ready_timeout=600)
    daft.set_execution_config(min_cpu_per_task=0)

    tos_dir_url = os.getenv("TOS_DIR_URL", "las-cn-beijing-public-online.tos-cn-beijing.volces.com")
    samples = {
        "image": [
            f"https://{tos_dir_url}/public/shared_image_dataset/cat_ip_adapter.jpeg"
        ]
    }

    image_src_type = "image_url"
    batch_size = 64
    model_path = os.getenv("MODEL_PATH", "/opt/las/models")
    model_name = "google/vit-base-patch16-224-in21k"
    dtype = "float32"
    use_cls_token_embedding = True
    rank = 0
    num_gpus = 1

    ds = daft.from_pydict(samples)
    ds = ds.with_column(
        "embedding",
        las_udf(
            ImageViTEmbedding,
            construct_args={
                "image_src_type": image_src_type,
                "batch_size": batch_size,
                "model_path": model_path,
                "model_name": model_name,
                "dtype": dtype,
                "use_cls_token_embedding": use_cls_token_embedding,
                "rank": rank,
            },
            num_gpus=num_gpus,
            batch_size=1,
        )(col("image")),
    )

    ds.show()

    # ╭────────────────────────────────┬────────────────────────────────╮
    # │ image                          ┆ embedding                      │
    # │ ---                            ┆ ---                            │
    # │ Utf8                           ┆ List[Float32]                  │
    # ╞════════════════════════════════╪════════════════════════════════╡
    # │ tos://las-cn-beijing-public-o… ┆[-0.011575016, -0.019808339, …  │
    # ╰────────────────────────────────┴────────────────────────────────╯