def __init__(self,
root=None,
split='train',
mode=None,
transform=None,
**kwargs):
root = root if root is not None else os.path.join(
dataset_dir(), 'COCO')
super(MSCOCO, self).__init__(root, split, mode, transform, **kwargs)
try_import_pycocotools()
from pycocotools.coco import COCO
from pycocotools import mask
if split == 'train':
print('train set')
ann_file = os.path.join(root,
'annotations/instances_train2017.json')
ids_file = os.path.join(root, 'annotations/train_ids.mx')
self.root = os.path.join(root, 'train2017')
else:
print('val set')
ann_file = os.path.join(root, 'annotations/instances_val2017.json')
ids_file = os.path.join(root, 'annotations/val_ids.mx')
self.root = os.path.join(root, 'val2017')
self.coco = COCO(ann_file)
self.coco_mask = mask
if os.path.exists(ids_file):
with open(ids_file, 'rb') as f:
self.ids = pickle.load(f)
else:
ids = list(self.coco.imgs.keys())
self.ids = self._preprocess(ids, ids_file)
self.transform = transform
def _load_jsons(self):
"""Load all image paths and labels from JSON annotation files into buffer."""
samples = dict()
labels = dict()
# lazy import pycocotools
try_import_pycocotools()
from pycocotools.coco import COCO
for split in self._splits:
anno = os.path.join(self.root, self.annotation_dir,
split) + '.json'
_coco = COCO(anno)
self._coco.append(_coco)
classes = [c['name'] for c in _coco.loadCats(_coco.getCatIds())]
if not classes == self.classes:
raise ValueError("Incompatible category names with COCO: ")
assert classes == self.classes
json_id_to_contiguous = {
v: k
for k, v in enumerate(_coco.getCatIds())
}
if self.json_id_to_contiguous is None:
self.json_id_to_contiguous = json_id_to_contiguous
self.contiguous_id_to_json = {
v: k
for k, v in self.json_id_to_contiguous.items()
}
else:
assert self.json_id_to_contiguous == json_id_to_contiguous
# iterate through the annotations
image_ids = sorted(_coco.getImgIds())
for entry in _coco.loadImgs(image_ids):
abs_path = self._parse_image_path(entry)
if not os.path.exists(abs_path):
raise IOError('Image: {} not exists.'.format(abs_path))
label = self._check_load_bbox(_coco, entry)
if not label:
continue
sample_id = len(samples)
samples[sample_id] = abs_path
labels[sample_id] = label
if sample_id not in self._im_shapes:
self._im_shapes[sample_id] = (entry['width'],
entry['height'])
return samples, labels
def _load_jsons(self):
"""Load all image paths and labels from JSON annotation files into buffer."""
images = dict()
boxes = dict()
captions = dict()
# lazy import pycocotools
try_import_pycocotools()
from pycocotools.coco import COCO
for split in self._splits:
anno = os.path.join(self.root, self.annotation_dir, split) + '.json'
_coco = COCO(anno)
self._coco.append(_coco)
anno = os.path.join(self.root, self.annotation_dir, split.replace('instances', 'captions')) + '.json'
_coco_cap = COCO(anno)
self._coco.append(_coco_cap)
categories = [c['name'] for c in _coco.loadCats(_coco.getCatIds())]
if not categories == self.categories:
raise ValueError("Incompatible category names with COCO: ")
assert categories == self.categories
json_id_to_contiguous = {
v: k for k, v in enumerate(_coco.getCatIds())}
if self.json_id_to_contiguous is None:
self.json_id_to_contiguous = json_id_to_contiguous
self.contiguous_id_to_json = {
v: k for k, v in self.json_id_to_contiguous.items()}
else:
assert self.json_id_to_contiguous == json_id_to_contiguous
# iterate through the annotations
for img_id, entry in zip(sorted(_coco.getImgIds()), _coco.loadImgs(sorted(_coco.getImgIds()))):
abs_path = self._parse_image_path(entry)
if not os.path.exists(abs_path):
raise IOError('Image: {} not exists.'.format(abs_path))
box = self._check_load_bbox(_coco, entry)
caption = self._load_captions(_coco_cap, entry)
if not box and not caption:
print("%s doesn't have any boxes or captions" % img_id)
continue
images[img_id] = abs_path
boxes[img_id] = box
captions[img_id] = caption
return images, boxes, captions
def _load_data(self):
"""Load all image paths and labels from JSON annotation files into buffer."""
items = []
labels = []
im_aspect_ratios = []
# lazy import pycocotools
try_import_pycocotools()
from pycocotools.coco import COCO
convert_gt_to_coco(self.data_path, self.channel, self.image_path,
self.field_name, self.annotation_path)
anno = os.path.join(self.data_path, self.annotation_path)
_coco = COCO(anno)
self._coco.append(_coco)
classes = [c['name'] for c in _coco.loadCats(_coco.getCatIds())]
if not classes == self.classes:
raise ValueError("Incompatible category names with COCO: ")
assert classes == self.classes
json_id_to_contiguous = {v: k for k, v in enumerate(_coco.getCatIds())}
if self.json_id_to_contiguous is None:
self.json_id_to_contiguous = json_id_to_contiguous
self.contiguous_id_to_json = {
v: k
for k, v in self.json_id_to_contiguous.items()
}
else:
assert self.json_id_to_contiguous == json_id_to_contiguous
# iterate through the annotations
image_ids = sorted(_coco.getImgIds())
for entry in _coco.loadImgs(image_ids):
abs_path = self._parse_image_path(entry)
if not os.path.exists(abs_path):
raise IOError('Image: {} not exists.'.format(abs_path))
label = self._check_load_bbox(_coco, entry)
if not label:
continue
im_aspect_ratios.append(float(entry['width']) / entry['height'])
items.append(abs_path)
labels.append(label)
return items, labels, im_aspect_ratios
def _update(self):
"""Use coco to get real scores. """
if not self._current_id == len(self._img_ids):
warnings.warn(
'Recorded {} out of {} validation images, incomplete results'.
format(self._current_id, len(self._img_ids)))
if not self._results:
# in case of empty results, push a dummy result
self._results.append({
'image_id': self._img_ids[0],
'category_id': 0,
'bbox': [0, 0, 0, 0],
'score': 0
})
import json
try:
with open(self._filename, 'w') as f:
json.dump(self._results, f)
except IOError as e:
raise RuntimeError(
"Unable to dump json file, ignored. What(): {}".format(str(e)))
# lazy import pycocotools
try_import_pycocotools()
if not hasattr(
self.dataset, 'coco'
): # it doesn't have a coco, probably isn't a coco set, so let us make a coco
json_path = self.dataset.build_coco_json()
from pycocotools.coco import COCO
self.dataset.coco = COCO(json_path)
pred = self.dataset.coco.loadRes(self._filename)
gt = self.dataset.coco
from pycocotools.cocoeval import COCOeval
coco_eval = COCOeval(gt, pred, 'bbox')
# coco_eval.params.recThrs = np.linspace(.0, 1.00, np.round((1.00 - .0) / .1) + 1, endpoint=True) ## tried to get the same as voc but still not same
coco_eval.evaluate()
coco_eval.accumulate()
self._coco_eval = coco_eval
return coco_eval
for url, checksum in _DOWNLOAD_URLS:
filename = download(url,
path=path,
overwrite=overwrite,
sha1_hash=checksum)
# extract
with zipfile.ZipFile(filename) as zf:
zf.extractall(path=path)
if __name__ == '__main__':
args = parse_args()
path = os.path.expanduser(args.download_dir)
if not os.path.isdir(path) or not os.path.isdir(os.path.join(path, 'train2017')) \
or not os.path.isdir(os.path.join(path, 'val2017')) \
or not os.path.isdir(os.path.join(path, 'annotations')):
if args.no_download:
raise ValueError(
('{} is not a valid directory, make sure it is present.'
' Or you should not disable "--no-download" to grab it'.
format(path)))
else:
download_coco(path, overwrite=args.overwrite)
# make symlink
makedirs(os.path.expanduser('~/.mxnet/datasets'))
if os.path.isdir(_TARGET_DIR):
os.remove(_TARGET_DIR)
os.symlink(path, _TARGET_DIR)
try_import_pycocotools()
# '46cdcf715b6b4f67e980b529534e79c2edffe084'),
# test2017.zip, for those who want to attend the competition.
# ('http://images.cocodataset.org/zips/test2017.zip',
# '4e443f8a2eca6b1dac8a6c57641b67dd40621a49'),
]
makedirs(path)
for url, checksum in _DOWNLOAD_URLS:
filename = download(url, path=path, overwrite=overwrite, sha1_hash=checksum)
# extract
with zipfile.ZipFile(filename) as zf:
zf.extractall(path=path)
if __name__ == '__main__':
args = parse_args()
path = os.path.expanduser(args.download_dir)
if not os.path.isdir(path) or not os.path.isdir(os.path.join(path, 'train2017')) \
or not os.path.isdir(os.path.join(path, 'val2017')) \
or not os.path.isdir(os.path.join(path, 'annotations')):
if args.no_download:
raise ValueError(('{} is not a valid directory, make sure it is present.'
' Or you should not disable "--no-download" to grab it'.format(path)))
else:
download_coco(path, overwrite=args.overwrite)
# make symlink
makedirs(os.path.expanduser('~/.mxnet/datasets'))
if os.path.isdir(_TARGET_DIR):
os.remove(_TARGET_DIR)
os.symlink(path, _TARGET_DIR)
try_import_pycocotools()
