Tensor Messages

Tensor metadata types for ML pipelines. For the full tensor API with DLPack zero-copy, see Tensor API.

Usage

Tensor data flows through HORUS using the Image, PointCloud, and Tensor pool-backed types — not message structs. The tensor descriptor types here are metadata for describing tensor shapes and dtypes.

For ML inference, the typical pattern is:

import horus
import numpy as np

def inference_tick(node):
    img = node.recv("camera.rgb")
    if img is None:
        return

    # Zero-copy to numpy via DLPack
    frame = np.from_dlpack(img)  # (H, W, C) uint8

    # Run inference
    result = model(frame)

    node.send("predictions", result)

See Tensor API for TensorHandle, TensorPool, and DLPack patterns.

See Also