GRID Docs | General Robotics

Input images up to 1280 x 720 (width x height). Larger images are rejected with HTTP 413, not downscaled — resize first.

Parameters

Union[str, Image.Image, np.ndarray]

Left stereo image.

Union[str, Image.Image, np.ndarray]

Right stereo image.

ImageInput | None

Rectified [left | right] frame, with equal-width views and at most 720x1280 pixels per eye. Supply this instead of left_image and right_image.

bool

default:"False"

If true, return a dictionary containing “depth” (float32 HxW) and “left_image” (uint8 RGB HxWx3). The left image can be passed to any segmentation model.

Dict[str, Any]

Camera parameters: - “K”: 3x3 intrinsics for the rectified left view at input resolution - “baseline”: Stereo baseline in meters - “hiera”: Hierarchy level (0-2) - “valid_iters”: Number of valid iterations

float | None

Optional HTTP timeout.

Returns

np.ndarray | dict[str, np.ndarray]: Depends on return_left_image:

False (default): The depth array.

True: A dictionary with “depth” and “left_image” arrays.

Depth is float32 with shape (H, W), in meters. The optional left image is uint8 RGB with shape (H, W, 3), aligned with depth. Both use the input per-eye resolution; the server performs no resizing or rectification.

Example

from grid_cortex_client import CortexClient, ModelType
import numpy as np
from PIL import Image
client = CortexClient()
K = np.array([[525, 0, 320], [0, 525, 240], [0, 0, 1]], dtype=np.float32)
aux = {"K": K, "baseline": 0.1, "hiera": 0, "valid_iters": 32}
left_image = np.array(Image.open("left.jpg"))
right_image = np.array(Image.open("right.jpg"))
depth = client.run(ModelType.FOUNDATIONSTEREO, left_image=left_image, right_image=right_image, aux_args=aux)
print(depth.shape)  # (480, 640)

Overview

Cortex Models

foundationstereo

Parameters

Returns

Example

​Parameters

​Returns

​Example

Parameters

Returns

Example