lib/core/get_gt_perturbed_proposals.py [22:50]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
def _augment_gt_boxes_by_perturbation(unique_gt_boxes, im_width, im_height):

    num_gt = unique_gt_boxes.shape[0]
    num_rois = 1000
    rois = np.zeros((num_rois, 4), dtype=np.float32)
    cnt = 0
    for i in range(num_gt):
        box = unique_gt_boxes[i]
        box_width = box[2] - box[0] + 1
        box_height = box[3] - box[1] + 1
        x_offset_max = (box_width - 1) // 2
        y_offset_max = (box_height - 1) // 2
        for _ in range(num_rois // num_gt):
            x_min_offset = np.random.uniform(low=-x_offset_max, high=x_offset_max)
            y_min_offset = np.random.uniform(low=-y_offset_max, high=y_offset_max)
            x_max_offset = np.random.uniform(low=-x_offset_max, high=x_offset_max)
            y_max_offset = np.random.uniform(low=-y_offset_max, high=y_offset_max)

            new_x_min = min(max(np.round(box[0] + x_min_offset), 0), im_width - 1)
            new_y_min = min(max(np.round(box[1] + y_min_offset), 0), im_height - 1)
            new_x_max = min(max(np.round(box[2] + x_max_offset), 0), im_width - 1)
            new_y_max = min(max(np.round(box[3] + y_max_offset), 0), im_height - 1)

            new_box = np.array(
                [new_x_min, new_y_min, new_x_max, new_y_max]).astype(np.float32)
            rois[cnt] = new_box
            cnt += 1

    return rois
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



lib/roi_data/fast_rcnn_rel.py [241:269]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
def _augment_gt_boxes_by_perturbation(unique_gt_boxes, im_width, im_height):

    num_gt = unique_gt_boxes.shape[0]
    num_rois = 1000
    rois = np.zeros((num_rois, 4), dtype=np.float32)
    cnt = 0
    for i in range(num_gt):
        box = unique_gt_boxes[i]
        box_width = box[2] - box[0] + 1
        box_height = box[3] - box[1] + 1
        x_offset_max = (box_width - 1) // 2
        y_offset_max = (box_height - 1) // 2
        for _ in range(num_rois // num_gt):
            x_min_offset = np.random.uniform(low=-x_offset_max, high=x_offset_max)
            y_min_offset = np.random.uniform(low=-y_offset_max, high=y_offset_max)
            x_max_offset = np.random.uniform(low=-x_offset_max, high=x_offset_max)
            y_max_offset = np.random.uniform(low=-y_offset_max, high=y_offset_max)

            new_x_min = min(max(np.round(box[0] + x_min_offset), 0), im_width - 1)
            new_y_min = min(max(np.round(box[1] + y_min_offset), 0), im_height - 1)
            new_x_max = min(max(np.round(box[2] + x_max_offset), 0), im_width - 1)
            new_y_max = min(max(np.round(box[3] + y_max_offset), 0), im_height - 1)

            new_box = np.array(
                [new_x_min, new_y_min, new_x_max, new_y_max]).astype(np.float32)
            rois[cnt] = new_box
            cnt += 1

    return rois
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -