class DeepScoresV2Dataset(CocoDataset):
def __init__(self,
ann_file,
pipeline,
classes=None,
data_root=None,
img_prefix='',
seg_prefix=None,
proposal_file=None,
test_mode=False,
filter_empty_gt=True,
use_oriented_bboxes=True):
self.filter_empty_gt = filter_empty_gt
super(DeepScoresV2Dataset,
self).__init__(ann_file, pipeline, data_root, img_prefix,
seg_prefix, proposal_file, test_mode)
#self.CLASSES = self.get_classes(classes)
self.use_oriented_bboxes = use_oriented_bboxes
@classmethod
def get_classes(cls, classes=None):
"""Get class names of current dataset.
Args:
classes (Sequence[str] | str | None): If classes is None, use
default CLASSES defined by builtin dataset. If classes is a
string, take it as a file name. The file contains the name of
classes where each line contains one class name. If classes is
a tuple or list, override the CLASSES defined by the dataset.
Returns:
tuple[str] or list[str]: Names of categories of the dataset.
"""
if classes is None:
return cls.CLASSES
if isinstance(classes, str):
# take it as a file path
class_names = mmcv.list_from_file(classes)
elif isinstance(classes, (tuple, list)):
class_names = classes
else:
raise ValueError(f'Unsupported type {type(classes)} of classes.')
return class_names
def load_annotations(self, ann_file):
self.obb = OBBAnns(ann_file)
self.obb.load_annotations()
self.obb.set_annotation_set_filter(['deepscores'])
# self.obb.set_class_blacklist(["staff"])
self.cat_ids = list(self.obb.get_cats().keys())
self.cat2label = {cat_id: i for i, cat_id in enumerate(self.cat_ids)}
self.label2cat = {v: k for k, v in self.cat2label.items()}
self.CLASSES = tuple(
[v["name"] for (k, v) in self.obb.get_cats().items()])
self.img_ids = [id['id'] for id in self.obb.img_info]
return self.obb.img_info
def get_ann_info(self, idx):
return self._parse_ann_info(*self.obb.get_img_ann_pair(idxs=[idx]))
def _filter_imgs(self, min_size=32):
valid_inds = []
for i, img_info in enumerate(self.obb.img_info):
if self.filter_empty_gt and len(img_info['ann_ids']) == 0:
continue
if min(img_info['width'], img_info['height']) >= min_size:
valid_inds.append(i)
return valid_inds
def _parse_ann_info(self, img_info, ann_info):
img_info, ann_info = img_info[0], ann_info[0]
gt_bboxes = []
gt_labels = []
gt_bboxes_ignore = np.zeros((0, 8 if self.use_oriented_bboxes else 4),
dtype=np.float32)
for i, ann in ann_info.iterrows():
# we have no ignore feature
if ann['area'] <= 0:
continue
bbox = ann['o_bbox' if self.use_oriented_bboxes else 'a_bbox']
gt_bboxes.append(bbox)
gt_labels.append(self.cat2label[ann['cat_id'][0]])
gt_bboxes = np.array(gt_bboxes, dtype=np.float32)
gt_labels = np.array(gt_labels, dtype=np.int64)
ann = dict(bboxes=gt_bboxes,
labels=gt_labels,
bboxes_ignore=gt_bboxes_ignore,
masks=None,
seg_map=None)
return ann
def prepare_json_dict(self, results):
json_results = {"annotation_set": "deepscores", "proposals": []}
for idx in range(len(self)):
img_id = self.img_ids[idx]
result = results[idx]
for label in range(len(result)):
bboxes = result[label]
for i in range(bboxes.shape[0]):
data = dict()
data['img_id'] = img_id
if len(bboxes[i]) == 8:
data['bbox'] = [str(nr) for nr in bboxes[i]]
data['score'] = 1
else:
data['bbox'] = [str(nr) for nr in bboxes[i][0:-1]]
data['score'] = str(bboxes[i][-1])
data['cat_id'] = self.label2cat[label]
json_results["proposals"].append(data)
return json_results
def write_results_json(self, results, filename=None):
if filename is None:
filename = "deepscores_results.json"
json_results = self.prepare_json_dict(results)
with open(filename, "w") as fo:
json.dump(json_results, fo)
return filename
def evaluate(self,
results,
metric='bbox',
logger=None,
jsonfile_prefix=None,
classwise=True,
proposal_nums=(100, 300, 1000),
iou_thrs=np.arange(0.5, 0.96, 0.05),
average_thrs=False,
work_dir=None):
"""Evaluation in COCO protocol.
Args:
results (list): Testing results of the dataset.
metric (str | list[str]): Metrics to be evaluated.
logger (logging.Logger | str | None): Logger used for printing
related information during evaluation. Default: None.
jsonfile_prefix (str | None): The prefix of json files. It includes
the file path and the prefix of filename, e.g., "a/b/prefix".
If not specified, a temp file will be created. Default: None.
classwise (bool): Whether to evaluating the AP for each class.
proposal_nums (Sequence[int]): Proposal number used for evaluating
recalls, such as recall@100, recall@1000.
Default: (100, 300, 1000).
iou_thrs (Sequence[float]): IoU threshold used for evaluating
recalls. If set to a list, the average recall of all IoUs will
also be computed. Default: 0.5.
Returns:
dict[str: float]
"""
metrics = metric if isinstance(metric, list) else [metric]
allowed_metrics = ['bbox']
for metric in metrics:
if metric not in allowed_metrics:
raise KeyError(f'metric {metric} is not supported')
filename = self.write_results_json(results)
self.obb.load_proposals(filename)
metric_results = self.obb.calculate_metrics(iou_thrs=iou_thrs,
classwise=classwise,
average_thrs=average_thrs)
categories = self.obb.get_cats()
metric_results = {
categories[key]['name']: value
for (key, value) in metric_results.items()
}
# add occurences
occurences_by_class = self.obb.get_class_occurences()
for (key, value) in metric_results.items():
value.update(no_occurences=occurences_by_class[key])
if work_dir is not None:
import pickle
import os
out_file = os.path.join(work_dir, "dsv2_metrics.pkl")
pickle.dump(metric_results, open(out_file, 'wb'))
print(metric_results)
return metric_results
def main(image_set, single_scale=False):
# (1)
if image_set == 'train':
#dir_txt = os.path.join(dir_dataset, 'labelTxt', image_set)
out_dir_json = os.path.join(dir_dataset, 'annotations', image_set)
os.makedirs(out_dir_json, exist_ok=True)
#txt2json(dir_txt, out_dir_json)
o = OBBAnns(dir_dataset + 'deepscores_train.json')
o.load_annotations()
#print(o)
cats = o.get_cats()
img_idxs = [i for i in range(len(o.img_info))]
imgs, anns = o.get_img_ann_pair(idxs=img_idxs,
ann_set_filter="deepscores")
filenames = []
for img in anns:
objs = []
img_np = np.array(img)
for object_instance in img_np:
obj = dict()
coord = np.array(object_instance[1],
dtype=np.float32).reshape([4, 2])
bbox = cv.boxPoints(cv.minAreaRect(coord)).astype(
np.int).tolist()
obj['name'] = cats[object_instance[2][0]]['name']
obj['bbox'] = bbox
objs.append(obj)
if objs:
filename = o.get_imgs(ids=[int(img_np[0][4])])[0]['filename']
filenames.append(os.path.splitext(filename)[0])
json_filename = os.path.splitext(filename)[0] + '.json'
json.dump(objs,
open(os.path.join(out_dir_json, json_filename),
'wt'),
indent=2)
# Split images and json annotations in train and val files
out_dir_train = os.path.join(dir_dataset, 'images2', 'train')
out_dir_val = os.path.join(dir_dataset, 'images2', 'val')
out_dir_val_json = os.path.join(dir_dataset, 'annotations', 'val')
out_dir_test = os.path.join(dir_dataset, 'images2', 'test')
os.makedirs(out_dir_train, exist_ok=True)
os.makedirs(out_dir_val, exist_ok=True)
os.makedirs(out_dir_test, exist_ok=True)
os.makedirs(out_dir_val_json, exist_ok=True)
filenames_train, filenames_val = train_test_split(filenames,
test_size=272,
random_state=8)
for filename in os.listdir(os.path.join(dir_dataset, 'images')):
if os.path.splitext(filename)[0] in filenames_train:
shutil.move(os.path.join(dir_dataset, 'images', filename),
os.path.join(out_dir_train, filename))
elif os.path.splitext(filename)[0] in filenames_val:
shutil.move(os.path.join(dir_dataset, 'images', filename),
os.path.join(out_dir_val, filename))
shutil.move(
os.path.join(out_dir_json,
os.path.splitext(filename)[0] + '.json'),
os.path.join(out_dir_val_json,
os.path.splitext(filename)[0] + '.json'))
else:
shutil.move(os.path.join(dir_dataset, 'images', filename),
os.path.join(out_dir_test, filename))
os.rmdir(os.path.join(dir_dataset, 'images'))
os.rename(os.path.join(dir_dataset, 'images2'),
os.path.join(dir_dataset, 'images'))
# (2)
pairs = []
for filename in os.listdir(os.path.join(dir_dataset, 'images', image_set)):
anno = os.path.join(dir_dataset, 'annotations', image_set,
filename.replace('png', 'json'))
img = os.path.join(dir_dataset, 'images', image_set, filename)
if not os.path.exists(anno):
anno = None
pairs.append([img, anno])
overlap = 0.25
sizes = [768] if single_scale else [512, 768, 1024, 1536]
save_empty = image_set == 'test'
image_set = f"{image_set}-{sizes[0]}" if single_scale else image_set
out_dir_images = os.path.join(dir_dataset, 'images', f'{image_set}-crop')
out_dir_annos = os.path.join(dir_dataset, 'annotations',
f'{image_set}-crop')
cropper = Cropper(sizes, overlap)
cropper.crop_batch(pairs, out_dir_images, out_dir_annos, save_empty)
# (3)
pairs = []
for filename in os.listdir(out_dir_images):
img = os.path.join('images', f'{image_set}-crop', filename)
anno = None if image_set == 'test' else os.path.join(
'annotations', f'{image_set}-crop', filename.replace(
'jpg', 'json'))
pairs.append([img, anno])
out_dir = os.path.join(dir_dataset, 'image-sets')
os.makedirs(out_dir, exist_ok=True)
json.dump(pairs,
open(os.path.join(out_dir, f'{image_set}.json'), 'wt'),
indent=2)
class DeepScoresV2Dataset(CocoDataset):
def load_annotations(self, ann_file):
self.obb = OBBAnns(ann_file)
self.obb.load_annotations()
self.obb.set_annotation_set_filter(['deepscores'])
self.obb.set_class_blacklist(["staff"])
self.cat_ids = list(self.obb.get_cats().keys())
self.cat2label = {cat_id: i for i, cat_id in enumerate(self.cat_ids)}
self.label2cat = {v: k for k, v in self.cat2label.items()}
self.CLASSES = tuple(
[v["name"] for (k, v) in self.obb.get_cats().items()])
self.img_ids = [id['id'] for id in self.obb.img_info]
return self.obb.img_info
def get_ann_info(self, idx):
return self._parse_ann_info(*self.obb.get_img_ann_pair(idxs=[idx]))
def _filter_imgs(self, min_size=32):
valid_inds = []
for i, img_info in enumerate(self.obb.img_info):
if self.filter_empty_gt and len(img_info['ann_ids']) == 0:
continue
if min(img_info['width'], img_info['height']) >= min_size:
valid_inds.append(i)
return valid_inds
def _parse_ann_info(self, img_info, ann_info):
img_info, ann_info = img_info[0], ann_info[0]
gt_bboxes = []
gt_labels = []
gt_bboxes_ignore = np.zeros((0, 4), dtype=np.float32)
for i, ann in ann_info.iterrows():
# we have no ignore feature
if ann['area'] <= 0:
continue
bbox = ann['a_bbox']
gt_bboxes.append(bbox)
gt_labels.append(self.cat2label[ann['cat_id'][0]])
gt_bboxes = np.array(gt_bboxes, dtype=np.float32)
gt_labels = np.array(gt_labels, dtype=np.int64)
ann = dict(bboxes=gt_bboxes,
labels=gt_labels,
bboxes_ignore=gt_bboxes_ignore,
masks=None,
seg_map=None)
return ann
def prepare_json_dict(self, results):
json_results = {"annotation_set": "deepscores", "proposals": []}
for idx in range(len(self)):
img_id = self.img_ids[idx]
result = results[idx]
for label in range(len(result)):
bboxes = result[label]
for i in range(bboxes.shape[0]):
data = dict()
data['img_id'] = img_id
data['bbox'] = [str(nr) for nr in bboxes[i][0:-1]]
data['score'] = str(bboxes[i][-1])
data['cat_id'] = self.label2cat[label]
json_results["proposals"].append(data)
return json_results
def write_results_json(self, results, filename=None):
if filename is None:
filename = "deepscores_results.json"
json_results = self.prepare_json_dict(results)
with open(filename, "w") as fo:
json.dump(json_results, fo)
return filename
def evaluate(self,
results,
metric='bbox',
logger=None,
jsonfile_prefix=None,
classwise=True,
proposal_nums=(100, 300, 1000),
iou_thrs=np.arange(0.5, 0.96, 0.05),
average_thrs=False):
"""Evaluation in COCO protocol.
Args:
results (list): Testing results of the dataset.
metric (str | list[str]): Metrics to be evaluated.
logger (logging.Logger | str | None): Logger used for printing
related information during evaluation. Default: None.
jsonfile_prefix (str | None): The prefix of json files. It includes
the file path and the prefix of filename, e.g., "a/b/prefix".
If not specified, a temp file will be created. Default: None.
classwise (bool): Whether to evaluating the AP for each class.
proposal_nums (Sequence[int]): Proposal number used for evaluating
recalls, such as recall@100, recall@1000.
Default: (100, 300, 1000).
iou_thrs (Sequence[float]): IoU threshold used for evaluating
recalls. If set to a list, the average recall of all IoUs will
also be computed. Default: 0.5.
Returns:
dict[str: float]
"""
metrics = metric if isinstance(metric, list) else [metric]
allowed_metrics = ['bbox']
for metric in metrics:
if metric not in allowed_metrics:
raise KeyError(f'metric {metric} is not supported')
filename = self.write_results_json(results)
self.obb.load_proposals(filename)
metric_results = self.obb.calculate_metrics(iou_thrs=iou_thrs,
classwise=classwise,
average_thrs=average_thrs)
metric_results = {
self.CLASSES[self.cat2label[key]]: value
for (key, value) in metric_results.items()
}
# add occurences
occurences_by_class = self.obb.get_class_occurences()
for (key, value) in metric_results.items():
value.update(no_occurences=occurences_by_class[key])
if True:
import pickle
pickle.dump(metric_results,
open('evaluation_renamed_rcnn.pickle', 'wb'))
print(metric_results)
return metric_results