def __init__(self, project, image_set):
imdb.__init__(self, project + '_' + image_set)
# specific config options
self.config = {}
self._project = project
self._image_set = image_set
self._data_path = osp.join(cfg.DATA_DIR, '')
cats = load_json(os.path.join(self._data_path, 'cats.json'))
cats = dict({k: v['name'] for k, v in cats.items()})
self._classes = tuple(['__background__'] + list(cats.values()))
self._class_to_ind = dict(
list(zip(self.classes, list(range(self.num_classes)))))
self._class_to_cat_id = dict([(v, k) for k, v in cats.items()])
self._image_index = self._load_image_set_index()
# Default to roidb handler
self.set_proposal_method('gt')
self.competition_mode(False)
self._images = load_json(os.path.join(self._data_path, 'images.json'))
self._annos = load_json(
os.path.join(self._data_path, 'annotations.json'))
print('images: %d, annotations: %d, dataset: %d' %
(len(self._images), len(self._annos), len(self._image_index)))
annoted = list(self._annos.keys())
self._image_index = list(set(self._image_index).intersection(annoted))
def verify_annotation(data_dir):
cats = load_json(os.path.join(data_dir, 'cats.json'))
img_j = load_json(os.path.join(data_dir, 'images.json'))
anno_j = load_json(os.path.join(data_dir, 'annotations.json'))
font = cv2.FONT_HERSHEY_SIMPLEX
for k, v in img_j.items():
full_name = os.path.join(data_dir, v['folder'], v['file_name'])
img = cv2.imread(full_name)
for cat_id in cats.keys():
anno = anno_j[k]
assert anno is not None
if not cat_id in anno:
continue
bbs = anno[cat_id]['bbox']
for b in bbs:
x, y, w, h = b['x'], b['y'], b['w'], b['h']
cv2.rectangle(img, (x, y), (x + w, y + h), (255, 0, 0), 2)
cv2.putText(img, cats[cat_id]['name'], (x, y), font, 1, (255, 255, 0), 2)
cv2.imshow('check annotations', img)
k = cv2.waitKey(0)
if k == ord('q'):
return
def convert_to_pascal_from_coco(big_dir, annos_json, images_json, cats_json,
prefix, annos_dir):
cats = load_json(cats_json)
images = load_json(images_json)
annos = load_json(annos_json)
if not os.path.exists(annos_dir):
os.makedirs(annos_dir)
i = 0
for k, v in annos.items():
info = images[k]
file_name = info['file_name']
new_file_name_stem = '%s_%05d' % (prefix, int(k))
new_file_name = new_file_name_stem + os.path.splitext(file_name)[-1]
save_to_xml2(info, v, cats, new_file_name,
os.path.join(annos_dir, new_file_name_stem + '.xml'))
old_file = os.path.join(big_dir, file_name)
new_file = os.path.join(big_dir, new_file_name)
os.rename(old_file, new_file)
i += 1
print('total %d files converted' % (i, ))
def __init__(self, project, image_set):
imdb.__init__(self, project + '_' + image_set)
self._project = project
self._image_set = image_set
self._data_path = os.path.join(cfg.DATA_DIR, 'ds_mj')
cats = load_json(os.path.join(cfg.DATA_DIR, 'cats.json'))
cats = dict({k: v['name'] for k, v in cats.items()})
self._classes = tuple(['__background__'] + list(cats.values()))
self._class_to_ind = dict(
list(zip(self.classes, list(range(self.num_classes)))))
self._image_ext = '.png'
self._image_index = self._load_image_set_index()
self.set_proposal_method('gt')
self.config = {}
assert os.path.exists(self._data_path), \
'Path does not exist: {}'.format(self._data_path)
print()
cv2.imshow('pic browser', image)
k = cv2.waitKey(0)
if k == ord('q'):
return True
return False
def browse_all(file_dir, cats):
smalls = {}
for k, v in cats.items():
full_name = os.path.join(cfg.data_dir, v['folder'], v['file_name'])
img = cv2.imread(full_name, 0)
if img is not None:
smalls[k] = img
for k, v in smalls.items():
print('broswering', cats[k]['name'])
done = browse(file_dir, v)
if done:
break
if __name__ == '__main__':
cfg.data_dir = '../dataset'
# template = cv2.imread(os.path.join(cfg.data_dir, 'atomitems/b8.png'), 0)
# browse(os.path.join(cfg.data_dir, 'bigpics'), template)
cats = load_json(os.path.join(cfg.data_dir, 'cats.json'))
browse_all(os.path.join(cfg.data_dir, 'bigpics'), cats)
def load_cats(cats_json):
cats = load_json(cats_json)
cats = {k: v['name'] for k, v in cats.items()}
class_to_catid = dict([(v, k) for k, v in cats.items()])
return cats, class_to_catid
import os
import numpy as np
import cv2
from annotate import load_json
data_dir = '../dataset'
cats = load_json(os.path.join(data_dir, 'cats.json'))
cats = dict({k: v['name'] for k, v in cats.items()})
print('cat num:', len(cats))
annos = load_json(os.path.join(data_dir, 'annotations.json'))
print('annotation num:', len(annos))
images = load_json(os.path.join(data_dir, 'images.json'))
print('image num:', len(images))
ids = list(annos.keys())
test_ids = ids[:5]
print('following images will be tested')
print(test_ids)
print(annos[1])
for i in test_ids:
img_info = images[i]
full_file_name = os.path.join(data_dir, img_info['folder'],
img_info['file_name'])
image = cv2.imread(full_file_name)
print(full_file_name)
if image is None:
linewidth=3.,
alpha=0.5)))
patches.append(
ax.text(bbox[0] + 3, (bbox[3] + bbox[1]) / 2,
'{:s}'.format(class_name),
bbox=dict(facecolor='blue', alpha=0.4),
fontsize=14,
color='white'))
return patches
cfg.DATA_DIR = './dataset'
cfg.ANCHOR_SCALES = [2.2, 4.2, 7.0, 7.6]
cfg.ANCHOR_RATIOS = [0.75, 0.83, 1.18, 1.33]
cats = load_json(os.path.join(cfg.DATA_DIR, 'cats.json'))
cat_ids = list(cats.keys())
CLASSES = [cats[i]['name'] for i in sorted(cat_ids)]
CLASSES = tuple(['__background__'] + CLASSES)
print('number of classes:', len(CLASSES))
NETS = {
'vgg16': ('vgg16_faster_rcnn_iter_20000.ckpt', ),
'res101': ('res101_faster_rcnn_iter_110000.ckpt', )
}
DATASETS = {'mahjong': ('mj_train', )}
def recognize(sess, net, im):
"""Detect object classes in an image using pre-computed object proposals."""
