Function generate_random_segment_data

ext/spt/utils/instance.py:300–335

Generate multiple images with random ground truth and predicted instance and semantic segmentation data. To make the images realistic, and to ensure that the instances form a PARTITION of the image, we rely on voronoi cells. Besides, to encourage a realistic overlap between the predi

(
        num_img=2,
        num_gt_per_img=10,
        num_pred_per_img=14,
        num_classes=2,
        height=32,
        width=64,
        shift=5,
        random_pred_label=False,
        verbose=True)

def generate_random_segment_data(
        num_img=2,
        num_gt_per_img=10,
        num_pred_per_img=14,
        num_classes=2,
        height=32,
        width=64,
        shift=5,
        random_pred_label=False,
        verbose=True):
    """Generate multiple images with random ground truth and predicted
    instance and semantic segmentation data. To make the images
    realistic, and to ensure that the instances form a PARTITION of the
    image, we rely on voronoi cells. Besides, to encourage a realistic
    overlap between the predicted and target instances, the prediction
    cell centers are sampled near the target samples.
    """
    tm_data = []
    spt_data = []

    for i_img in range(num_img):
        if verbose:
            print(f"\nImage {i_img + 1}/{num_img}")
        tm_data_, spt_data_ = generate_single_random_segment_image(
            num_gt=num_gt_per_img,
            num_pred=num_pred_per_img,
            num_classes=num_classes,
            height=height,
            width=width,
            shift=shift,
            random_pred_label=random_pred_label,
            show=verbose)
        tm_data.append(tm_data_)
        spt_data.append(spt_data_)

    return tm_data, spt_data


def _instance_cut_pursuit(