def generate_output_image()

in sagemaker/src/word_and_line_segmentation.py [0:0]

• 48 lines of code
• 2 McCabe index (conditional complexity)

def generate_output_image(box_predictions, default_anchors, cls_probs, box_target, box_mask, cls_target, x, y):
    '''
    Generate the image with the predicted and actual bounding boxes.
    Parameters
    ----------
    box_predictions: nd.array
        Bounding box predictions relative to the anchor boxes, output of the network

    default_anchors: nd.array
        Anchors used, output of the network
    
    cls_probs: nd.array
        Output of nd.SoftmaxActivation(nd.transpose(class_predictions, (0, 2, 1)), mode='channel')
        where class_predictions is the output of the network.

    box_target: nd.array
        Output classification probabilities from network.training_targets(default_anchors, class_predictions, y)

    box_mask: nd.array
        Output bounding box predictions from network.training_targets(default_anchors, class_predictions, y) 

    cls_target: nd.array
        Output targets from network.training_targets(default_anchors, class_predictions, y)
    
    x: nd.array
       The input images

    y: nd.array
        The actual labels

    Returns
    -------
    output_image: np.array
        The images with the predicted and actual bounding boxes drawn on

    number_of_bbs: int
        The number of predicting bounding boxes
    '''
    output = MultiBoxDetection(*[cls_probs, box_predictions, default_anchors], force_suppress=True, clip=False)
    output = box_nms(output, overlap_thresh=overlap_thres, valid_thresh=min_c, topk=topk)
    output = output.asnumpy()

    number_of_bbs = 0
    predicted_bb = []
    for b in range(output.shape[0]):
        predicted_bb_ = output[b, output[b, :, 0] != -1]
        predicted_bb_ = predicted_bb_[:, 2:]
        number_of_bbs += predicted_bb_.shape[0]
        predicted_bb_[:, 2] = predicted_bb_[:, 2] - predicted_bb_[:, 0]
        predicted_bb_[:, 3] = predicted_bb_[:, 3] - predicted_bb_[:, 1]
        predicted_bb.append(predicted_bb_)
        
    labels = y[:, :, 1:].asnumpy()
    labels[:, :, 2] = labels[:, :, 2] - labels[:, :, 0]
    labels[:, :, 3] = labels[:, :, 3] - labels[:, :, 1]

    output_image = draw_boxes_on_image(predicted_bb, labels, x.asnumpy())
    output_image[output_image<0] = 0
    output_image[output_image>1] = 1

    return output_image, number_of_bbs