infer.py

Generate optical flow with one of the available models.

This script can display and save optical flow estimated by any of the available models. It accepts multiple types of inputs, including: individual images, a folder of images, a video, or a webcam stream.

infer.infer(args: Namespace, model: BaseModel) → None[source]

Perform the inference.

Parameters:

modelBaseModel: The model to be used for inference.
argsNamespace: Arguments to configure the model and the inference.

See also

ptlflow.models.base_model.base_model.BaseModel: The parent class of the available models.

infer.init_input(input_path: str | List[str]) → Tuple[VideoCapture, List[Path], int, ndarray][source]

Initialize the required variable to start loading the inputs.

This function will detect which type of input_path was given (list of images, folder of images, video, or webcam). Then it will establish its length and also get the first frame of the input.

Parameters:

input_pathstr: The path to the input(s).

Returns:

tuple[cv.VideoCapture, List[Path], int, np.ndarray]: The initialized variables - a cv.VideoCapture if the input is a video OR - a list of paths to the images otherwise, - the maximum number of images, and - the first image.

infer.show_outputs(img1: ndarray, img2: ndarray, preds_npy: Dict[str, ndarray], auto_forward: bool, max_show_side: int) → int[source]

Show the images on the screen.

Parameters:

img1np.ndarray: First image for estimating the optical flow.
img2np.ndarray: Second image for estimating the optical flow.
preds_npydict[str, np.ndarray]: The model predictions converted to numpy format.
auto_forwardbool: If false, the user needs to press a key to move to the next image.
max_show_sideint: If max(height, width) of the image is larger than this value, then it is downscaled before showing.

Returns:

int: A value representing which key the user pressed.