fix eval_instance_segmentation_coco when pred_masks is empty

[yuyu2172]

Why

When pred_mask.shape == (0, H, W), the function fails.

Note

The test data did not follow the current data format, so I fixed it with the following conversion script.

import numpy as np

data = np.load('eval_instance_segmentation_coco_result_2018_07_06.npz', encoding='latin1')

labels = data['labels']
masks = data['masks']
scores = data['scores']
masks = [np.array(mask) for mask in masks]

np.savez(
    'eval_instance_segmentation_coco_result_2019_02_12.npz',
    labels=labels,
    scores=scores,
    masks=masks)

[yuyu2172]

In the original data, masks are stored as [[(H, W)]].
This contradicts with the format of masks, which is [(R, H, W)].