def run(args):
dataset = VOCInstanceSegmentationDataset(split=args.chainer_eval_set,
data_dir=args.data_root)
gt_masks = [
dataset.get_example_by_keys(i, (1, ))[0] for i in range(len(dataset))
]
gt_labels = [
dataset.get_example_by_keys(i, (2, ))[0] for i in range(len(dataset))
]
pred_class = []
pred_mask = []
pred_score = []
for id in dataset.ids:
ins_out = np.load(os.path.join(args.ins_seg_out_dir, id + '.npy'),
allow_pickle=True).item()
pred_class.append(ins_out['class'])
pred_mask.append(ins_out['mask'])
pred_score.append(ins_out['score'])
print(
'0.5iou:',
chainercv.evaluations.eval_instance_segmentation_voc(pred_mask,
pred_class,
pred_score,
gt_masks,
gt_labels,
iou_thresh=0.5))
def run(args):
dataset = VOCInstanceSegmentationDataset(split=args.chainer_eval_set,
data_dir=args.voc12_root)
pred_class = []
pred_mask = []
pred_score = []
gt_masks = []
gt_labels = []
out_dir = None
n_img = 0
for i, id in enumerate(dataset.ids):
print(id)
if (out_dir is not None) and (not os.path.exists(
os.path.join(out_dir, id + '.npy'))):
continue
gt_masks.append(dataset.get_example_by_keys(i, (1, ))[0])
gt_labels.append(dataset.get_example_by_keys(i, (2, ))[0])
ins_out = np.load(os.path.join(args.ins_seg_out_dir, id + '.npy'),
allow_pickle=True).item()
pred_class.append(ins_out['class'])
pred_mask.append(ins_out['mask'])
pred_score.append(ins_out['score'])
n_img += 1
print(n_img)
print(
'0.5iou:',
chainercv.evaluations.eval_instance_segmentation_voc(pred_mask,
pred_class,
pred_score,
gt_masks,
gt_labels,
iou_thresh=0.5))
print(
'0.75iou:',
chainercv.evaluations.eval_instance_segmentation_voc(pred_mask,
pred_class,
pred_score,
gt_masks,
gt_labels,
iou_thresh=0.75))
def setUp(self):
self.dataset = VOCInstanceSegmentationDataset(split=self.split)