def convert_keypoints(fname, prefix, subset=None):
regions_json = []
keypoints_train_data = json.load(file(fname))
kp_id_to_category = {
k['id']: k
for k in keypoints_train_data['categories']
}
for annotation in keypoints_train_data['annotations']:
if subset is None or annotation['image_id'] in subset:
annotation['category'] = kp_id_to_category[
annotation['category_id']]
x = annotation['bbox'][0]
y = annotation['bbox'][1]
w = annotation['bbox'][2]
h = annotation['bbox'][3]
fname = "{}{}.jpg".format(prefix,
str(annotation['image_id']).zfill(12))
a = annotation
object_name = 'coco_keypoints/{}/{}'.format(
a[u'category'][u'supercategory'], a[u'category'][u'name'])
regions_json.append(
utils.create_region_json(fname,
x=x,
y=y,
w=w,
h=h,
object_name=object_name,
metadata=annotation,
text=None))
return regions_json
def convert_instances(fname, prefix, subset=None):
regions_json = []
data = defaultdict(lambda: {
'image': None,
'annotations': [],
'captions': [],
'keypoints': []
})
train_data = json.load(file(fname))
id_to_license = {k['id']: k for k in train_data['licenses']}
id_to_category = {k['id']: k for k in train_data['categories']}
for entry in train_data['images']:
if subset is None or entry['id'] in subset:
entry['license'] = id_to_license[entry['license']]
data[entry['id']]['image'] = entry
fname = "{}{}.jpg".format(prefix, str(entry['id']).zfill(12))
regions_json.append(
utils.create_region_json(fname,
"info",
0,
0,
0,
0,
entry,
None,
full_frame=True))
for a in train_data['annotations']:
if subset is None or a['image_id'] in subset:
a['category'] = id_to_category[a['category_id']]
x = a['bbox'][0]
y = a['bbox'][1]
w = a['bbox'][2]
h = a['bbox'][3]
object_name = 'coco_instance/{}/{}'.format(
a[u'category'][u'supercategory'], a[u'category'][u'name'])
fname = "{}{}.jpg".format(prefix, str(a['image_id']).zfill(12))
regions_json.append(
utils.create_region_json(fname,
x=x,
y=y,
w=w,
h=h,
object_name=object_name,
metadata=a,
text=None))
return regions_json
def convert_captions(fname,prefix,subset=None):
regions_json = []
captions_train_data = json.load(file(fname))
for annotation in captions_train_data['annotations']:
if subset is None or annotation['image_id'] in subset:
fname = "{}{}.jpg".format(prefix,str(annotation['image_id']).zfill(12))
regions_json.append(utils.create_region_json(fname,"caption",0,0,0,0,None,annotation['caption'],full_frame=True))
return regions_json