Operator ID: daft.las.functions.video.video_adaptive_keyframes_sampling.VideoAdaptiveKeyframesSampling

Description #

The video adaptive keyframe extraction processor uses CLIP and adaptive algorithms for intelligent frame selection.

Key features: #

• Semantic understanding: Calculates the similarity between video frames and text descriptions based on the CLIP model, intelligently selecting representative keyframes.
• Multi-format support: Compatible with local paths/URLs, base64 encoding, and binary stream input.
• Efficient processing: Supports batch inference and distributed computing, significantly improving processing speed.
• Flexible configuration: Allows customization of the CLIP model and inference precision to meet different scenario requirements.

Application scenarios: #

• Video content summarization, video retrieval, video analysis, and multimedia content management.

Operator parameters #

Input #

Output #

An array of structs containing keyframe information. Each element includes:

• tos_paths: List of TOS paths for keyframe images
• frame_ids: List of original frame indices for keyframes
• scores: List of semantic similarity scores for keyframes

Parameters #

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

Examples #

The following code demonstrates how to use Daft (for distributed scenarios) to run the operator for video adaptive keyframe extraction.

from __future__ import annotations

import os

import daft
from daft import col
from daft.las.functions.udf import las_udf
from daft.las.functions.video.video_adaptive_keyframes_sampling import VideoAdaptiveKeyframesSampling

if __name__ == "__main__":
    # The extracted keyframes will be saved to the specified TOS path. Therefore, you need to set environment variables to ensure write permissions to TOS, including: ACCESS_KEY, SECRET_KEY, TOS_ENDPOINT, TOS_REGION, TOS_TEST_DIR
    TOS_TEST_DIR = os.getenv("TOS_TEST_DIR", "your-bucket")
    output_tos_dir = f"tos://{TOS_TEST_DIR}/video_adaptive_keyframes_sampling"
    
    if os.getenv("DAFT_RUNNER", "native") == "ray":
        import logging

        import 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",
            )
            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)

        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)

    # Construct the URL using environment variables
    tos_dir_url = os.getenv("TOS_DIR_URL", "las-cn-beijing-public-online.tos-cn-beijing.volces.com")
    samples = {
        "video": [
            f"https://{tos_dir_url}/public/shared_video_dataset/cooking.mp4"
        ]
    }

    ds = daft.from_pydict(samples)

    keyframe_sampler = las_udf(
        VideoAdaptiveKeyframesSampling,
        construct_args={
            "text": "spoon",
            "output_tos_dir": output_tos_dir,
            "max_num_frames": 4,
            "fps": 2.0,
            "batch_size": 16,
            "rank": 0,
            "return_keyframes": True,
            "img_type": ".jpg", 
        },
        num_gpus=1,
        concurrency=1,
        batch_size=1,
    )

    # Use Daft for distributed processing
    ds = ds.with_column("keyframes", keyframe_sampler(col("video")))

    ds.show()
    # ╭────────────────────────────────┬───────────────────────────────────────────────────────────────────────────────╮
    # │ video                          ┆  keyframes                                                                    │
    # │ ---                            ┆ ---                                                                           │
    # │ Utf8                           ┆ Struct[tos_paths: List[Utf8], frame_ids: List[Int64], scores: List[Float32]]  │
    # ╞════════════════════════════════╪═══════════════════════════════════════════════════════════════════════════════╡
    # │ https://las-cn-beijing-publi-… ┆  {tos_paths: [las-cn-beijing…, ...], frame_ids: [123, ...], scores: [...]}    │
    # ╰────────────────────────────────┴───────────────────────────────────────────────────────────────────────────────╯