def main():
from common import get_train_imgpaths
import jsk_apc2015_common
rospy.init_node('extract_color_histogram')
all_objects = jsk_apc2015_common.get_object_list()
color_space_param = rospy.get_param('~color_space', 'lab')
if color_space_param == 'rgb':
colors = ['red', 'green', 'blue']
elif color_space_param == 'lab':
colors = ['l']
else:
raise ValueError('Unknown color space')
rospy.loginfo('color space: {c}'.format(c=color_space_param))
object_param = rospy.get_param('~object', all_objects)
object_nms = object_param.split(',')
if len(object_nms) == 1 and object_nms[0] == 'all':
object_nms = all_objects
rospy.loginfo('objects: {obj}'.format(obj=object_nms))
for object_nm in object_nms:
if object_nm not in all_objects:
rospy.logwarn('Unknown object, skipping: {}'.format(object_nm))
else:
imgpaths = get_train_imgpaths(object_nm)
raw_paths, mask_paths = zip(*imgpaths)
for color in colors:
e = ExtractColorHistogram(object_nm=object_nm, color=color,
raw_paths=raw_paths, mask_paths=mask_paths)
e.extract_and_save()
def main():
parser = argparse.ArgumentParser()
parser.add_argument("bof_histogram")
args = parser.parse_args(sys.argv[1:])
print("loading bof histogram")
with gzip.open(args.bof_histogram, "rb") as f:
obj_hists = pickle.load(f)
target_names = jsk_apc2015_common.get_object_list()
# create train and test data
X, y = [], []
for i, obj_name in enumerate(target_names):
X.append(obj_hists[obj_name])
y += [i] * len(obj_hists[obj_name])
X = np.vstack(X)
normalize(X, copy=False)
y = np.array(y)
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=np.random.randint(1234))
# train and test
lgr = LogisticRegression()
print("fitting LogisticRegression")
lgr.fit(X_train, y_train)
with gzip.open("lgr.pkl.gz", "wb") as f:
pickle.dump(lgr, f)
y_pred = lgr.predict(X_test)
print("score lgr: {}".format(accuracy_score(y_test, y_pred)))
print(classification_report(y_test, y_pred, target_names=target_names))
def get_object_sizes(data_dir):
cache_file = 'object_sizes.pkl'
if osp.exists(cache_file):
return pickle.load(open(cache_file, 'rb'))
img_shape = None
objects = jsk_apc2015_common.get_object_list()
df = []
for obj in objects:
mask_files = os.listdir(osp.join(data_dir, obj, 'masks'))
for f in mask_files:
if f.startswith('NP'):
continue
mask = cv2.imread(osp.join(data_dir, obj, 'masks', f), 0)
if img_shape is None:
img_shape = mask.shape
else:
assert img_shape == mask.shape
mask = (mask > 127).astype(int)
size = mask.sum()
df.append([objects.index(obj), obj, f, size])
df = pd.DataFrame(df)
df.columns = ['object_index', 'object_name', 'fname', 'size']
pickle.dump(df, open(cache_file, 'wb'))
return df
def main():
from common import get_train_imgpaths
import jsk_apc2015_common
rospy.init_node('extract_color_histogram')
all_objects = jsk_apc2015_common.get_object_list()
color_space_param = rospy.get_param('~color_space', 'lab')
if color_space_param == 'rgb':
colors = ['red', 'green', 'blue']
elif color_space_param == 'lab':
colors = ['l']
else:
raise ValueError('Unknown color space')
rospy.loginfo('color space: {c}'.format(c=color_space_param))
object_param = rospy.get_param('~object', all_objects)
object_nms = object_param.split(',')
if len(object_nms) == 1 and object_nms[0] == 'all':
object_nms = all_objects
rospy.loginfo('objects: {obj}'.format(obj=object_nms))
for object_nm in object_nms:
if object_nm not in all_objects:
rospy.logwarn('Unknown object, skipping: {}'.format(object_nm))
else:
imgpaths = get_train_imgpaths(object_nm)
raw_paths, mask_paths = zip(*imgpaths)
for color in colors:
e = ExtractColorHistogram(object_nm=object_nm,
color=color,
raw_paths=raw_paths,
mask_paths=mask_paths)
e.extract_and_save()
def _predict(self, img_msg, label_msg):
# convert image
bridge = cv_bridge.CvBridge()
input_image = bridge.imgmsg_to_cv2(img_msg, 'rgb8')
input_label = bridge.imgmsg_to_cv2(label_msg)
# predict
region_imgs = []
for l in np.unique(input_label):
if l == 0: # bg_label
continue
mask = (input_label == l)
region = jsk_recognition_utils.bounding_rect_of_mask(
input_image, mask)
region_imgs.append(region)
y_proba = self.estimator.predict(region_imgs)
target_names = np.array(jsk_apc2015_common.get_object_list())
y_pred = np.argmax(y_proba, axis=-1)
label_proba = [p[i] for p, i in zip(y_proba, y_pred)]
# prepare message
res = ClassificationResult()
res.header = img_msg.header
res.labels = y_pred
res.label_names = target_names[y_pred]
res.label_proba = label_proba
res.probabilities = y_proba.reshape(-1)
res.classifier = '<jsk_2015_05_baxter_apc.ColorHistogramFeatures>'
res.target_names = target_names
self._pub.publish(res)
def main():
rospy.init_node('extract_sift')
obj_names = jsk_apc2015_common.get_object_list()
for obj_name in obj_names:
if load_siftdata(obj_name, dry_run=True):
continue # already extracted
extract_sift(obj_name)
def visualize_stow_contents(work_order):
from jsk_apc2015_common.util import rescale
tote_img = cv2.imread(osp.join(PKG_PATH, 'models/tote/image.jpg'))
object_list = jsk_apc2015_common.get_object_list()
object_imgs = {}
for obj in object_list:
img_path = osp.join(OLD_PKG_PATH, 'models/{obj}/image.jpg'.format(obj=obj))
img = cv2.imread(img_path)
h, w = img.shape[:2]
if h > w:
img = np.rollaxis(img, 1)
object_imgs[obj] = img
# draw object images on tote image
tote_region = [[190,230],[1080,790]]
region_h = tote_region[1][1] - tote_region[0][1]
region_w = tote_region[1][0] - tote_region[0][0]
max_obj_h,max_obj_w = region_h / 3, region_w / 4
tote_x_min,tote_y_min = tote_region[0][0], tote_region[0][1]
x_min, y_min = tote_x_min, tote_y_min
for obj in work_order:
obj_img = object_imgs[obj]
scale_h = 1. * max_obj_h / obj_img.shape[0]
scale_w = 1. * max_obj_w / obj_img.shape[1]
scale = min([scale_h, scale_w])
obj_img = rescale(obj_img, scale)
obj_h, obj_w = obj_img.shape[:2]
x_max, y_max = x_min + obj_w, y_min + obj_h
tote_img[y_min:y_max, x_min:x_max] = obj_img
x_min += max_obj_w
if x_max >= region_w :
x_min = tote_x_min
y_min += max_obj_h
return tote_img
def spin_once(self):
if self.bof_data is None or self.cfeature is None:
return
stamp, bof_objects_proba = self.bof_data
stamp, cfeature_objects_proba = self.cfeature
weight = self.weight
target_bin = rospy.get_param('target_bin', None)
object_list = jsk_apc2015_common.get_object_list()
all_proba = [
(o,
(weight[o]['bof'] * bof_objects_proba[o]) +
(weight[o]['color'] * cfeature_objects_proba[o])
) for o in object_list
]
# verification result for debug
candidates = self.bin_contents.get(target_bin, None)
if candidates is None:
candidates = object_list
matched = sorted(all_proba, key=lambda x: x[1])[-1][0]
# compose msg
msg = ObjectRecognition()
msg.header.stamp = stamp
msg.matched = matched
msg.probability = dict(all_proba)[matched] / sum(dict(all_proba).values())
msg.candidates = candidates
msg.probabilities = np.array([dict(all_proba)[c] for c in candidates])
msg.probabilities /= msg.probabilities.sum()
self.pub_debug.publish(msg)
# verification result with json target
if target_bin is None or target_bin == '':
return
proba = [
(c,
(weight[c]['bof'] * bof_objects_proba[c]) +
(weight[c]['color'] * cfeature_objects_proba[c])
) for c in candidates
]
matched = sorted(proba, key=lambda x: x[1])[-1][0]
# compose msg
msg = ObjectRecognition()
msg.header.stamp = stamp
msg.matched = matched
msg.probability = dict(proba)[matched] / sum(dict(proba).values())
msg.candidates = candidates
msg.probabilities = np.array([dict(proba)[c] for c in candidates])
msg.probabilities /= msg.probabilities.sum()
self.pub.publish(msg)
def get_sift_descriptors(n_imgs=None, data_dir=None):
objects = jsk_apc2015_common.get_object_list()
obj_descs = []
for obj in objects:
descs = load_siftdata(obj_name=obj,
return_pos=False,
data_dir=data_dir)
if descs is None:
continue
if n_imgs is None:
n_imgs = len(descs)
p = np.random.randint(0, len(descs), size=n_imgs)
descs = np.array(map(lambda x: x.astype('float16'), descs))
obj_descs.append((obj, descs[p]))
return obj_descs
def spin_once(self):
if self.bof_data is None or self.cfeature is None:
return
stamp, bof_objects_proba = self.bof_data
stamp, cfeature_objects_proba = self.cfeature
weight = self.weight
target_bin = rospy.get_param('target_bin', None)
object_list = jsk_apc2015_common.get_object_list()
all_proba = [(o, (weight[o]['bof'] * bof_objects_proba[o]) +
(weight[o]['color'] * cfeature_objects_proba[o]))
for o in object_list]
# verification result for debug
candidates = self.bin_contents.get(target_bin, None)
if candidates is None:
candidates = object_list
matched = sorted(all_proba, key=lambda x: x[1])[-1][0]
# compose msg
msg = ObjectRecognition()
msg.header.stamp = stamp
msg.matched = matched
msg.probability = dict(all_proba)[matched] / sum(
dict(all_proba).values())
msg.candidates = candidates
msg.probabilities = np.array([dict(all_proba)[c] for c in candidates])
msg.probabilities /= msg.probabilities.sum()
self.pub_debug.publish(msg)
# verification result with json target
if target_bin is None or target_bin == '':
return
proba = [(c, (weight[c]['bof'] * bof_objects_proba[c]) +
(weight[c]['color'] * cfeature_objects_proba[c]))
for c in candidates]
matched = sorted(proba, key=lambda x: x[1])[-1][0]
# compose msg
msg = ObjectRecognition()
msg.header.stamp = stamp
msg.matched = matched
msg.probability = dict(proba)[matched] / sum(dict(proba).values())
msg.candidates = candidates
msg.probabilities = np.array([dict(proba)[c] for c in candidates])
msg.probabilities /= msg.probabilities.sum()
self.pub.publish(msg)
def visualize_stow_contents(work_order):
"""Visualize stow contents with passed work order.
Args:
work_order (list): objects in the stow.
Returns:
tote_img (~numpy.ndarray): image of objects over the tote.
"""
from jsk_apc2015_common.util import rescale
rp = rospkg.RosPack()
pkg_path = rp.get_path(PKG)
tote_img = cv2.imread(osp.join(pkg_path, 'models/tote/image.jpg'))
object_list = jsk_apc2015_common.get_object_list()
object_imgs = {}
pkg_path = rp.get_path('jsk_apc2015_common')
for obj in object_list:
img_path = osp.join(pkg_path, 'models/{obj}/image.jpg'.format(obj=obj))
img = cv2.imread(img_path)
h, w = img.shape[:2]
if h > w:
img = np.rollaxis(img, 1)
object_imgs[obj] = img
# draw object images on tote image
tote_region = [[190, 230], [1080, 790]]
region_h = tote_region[1][1] - tote_region[0][1]
region_w = tote_region[1][0] - tote_region[0][0]
max_obj_h, max_obj_w = region_h / 3, region_w / 4
tote_x_min, tote_y_min = tote_region[0][0], tote_region[0][1]
x_min, y_min = tote_x_min, tote_y_min
for obj in work_order:
obj_img = object_imgs[obj]
scale_h = 1. * max_obj_h / obj_img.shape[0]
scale_w = 1. * max_obj_w / obj_img.shape[1]
scale = min([scale_h, scale_w])
obj_img = rescale(obj_img, scale)
obj_h, obj_w = obj_img.shape[:2]
x_max, y_max = x_min + obj_w, y_min + obj_h
tote_img[y_min:y_max, x_min:x_max] = obj_img
x_min += max_obj_w
if x_max >= region_w:
x_min = tote_x_min
y_min += max_obj_h
return tote_img
def visualize_stow_contents(work_order):
"""Visualize stow contents with passed work order.
Args:
work_order (list): objects in the stow.
Returns:
tote_img (~numpy.ndarray): image of objects over the tote.
"""
from jsk_apc2015_common.util import rescale
rp = rospkg.RosPack()
pkg_path = rp.get_path(PKG)
tote_img = cv2.imread(osp.join(pkg_path, 'models/tote/image.jpg'))
object_list = jsk_apc2015_common.get_object_list()
object_imgs = {}
pkg_path = rp.get_path('jsk_apc2015_common')
for obj in object_list:
img_path = osp.join(pkg_path, 'models/{obj}/image.jpg'.format(obj=obj))
img = cv2.imread(img_path)
h, w = img.shape[:2]
if h > w:
img = np.rollaxis(img, 1)
object_imgs[obj] = img
# draw object images on tote image
tote_region = [[190, 230], [1080, 790]]
region_h = tote_region[1][1] - tote_region[0][1]
region_w = tote_region[1][0] - tote_region[0][0]
max_obj_h, max_obj_w = region_h / 3, region_w / 4
tote_x_min, tote_y_min = tote_region[0][0], tote_region[0][1]
x_min, y_min = tote_x_min, tote_y_min
for obj in work_order:
obj_img = object_imgs[obj]
scale_h = 1. * max_obj_h / obj_img.shape[0]
scale_w = 1. * max_obj_w / obj_img.shape[1]
scale = min([scale_h, scale_w])
obj_img = rescale(obj_img, scale)
obj_h, obj_w = obj_img.shape[:2]
x_max, y_max = x_min + obj_w, y_min + obj_h
tote_img[y_min:y_max, x_min:x_max] = obj_img
x_min += max_obj_w
if x_max >= region_w:
x_min = tote_x_min
y_min += max_obj_h
return tote_img
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--year', type=int, default=2016)
args = parser.parse_args()
if args.year == 2015:
import jsk_apc2015_common
cls_names = ['background'] + jsk_apc2015_common.get_object_list()
elif args.year == 2016:
import jsk_apc2016_common
data = jsk_apc2016_common.get_object_data()
cls_names = ['background'] + [d['name'] for d in data]
else:
raise ValueError
text = []
for cls_id, cls_name in enumerate(cls_names):
text.append('{:2}: {}'.format(cls_id, cls_name))
print('\n'.join(text))
parser = argparse.ArgumentParser()
parser.add_argument('json_id')
args = parser.parse_args()
json_id = args.json_id
N = 2
target_bin = 'h'
abandon_objects = [
'genuine_joe_plastic_stir_sticks',
'cheezit_big_original',
'rolodex_jumbo_pencil_cup',
'champion_copper_plus_spark_plug',
'oreo_mega_stuf',
]
objects = jsk_apc2015_common.get_object_list()
target_obj = None
while (target_obj is None) or (target_obj in abandon_objects):
candidates = []
for i in xrange(N):
i_obj = np.random.randint(0, len(objects))
candidates.append(objects[i_obj])
i_target = np.random.randint(0, len(candidates))
target_obj = candidates[i_target]
json_data = {
'bin_contents': {
'bin_{0}'.format(target_bin.upper()): candidates,
},
'work_order': [
def __init__(self):
SiftMatcher.__init__(self)
ObjectMatcher.__init__(self, '/semi/sift_matcher')
self.object_list = jsk_apc2015_common.get_object_list()
self.siftdata_cache = {}
def __init__(self):
SiftMatcher.__init__(self)
ObjectMatcher.__init__(self, '/semi/sift_matcher')
self.object_list = jsk_apc2015_common.get_object_list()
self.siftdata_cache = {}
def test_get_object_list():
objects = jsk_apc2015_common.get_object_list()
assert_equal(25, len(objects))
parser = argparse.ArgumentParser()
parser.add_argument('json_id')
args = parser.parse_args()
json_id = args.json_id
N = 2
target_bin = 'h'
abandon_objects = [
'genuine_joe_plastic_stir_sticks',
'cheezit_big_original',
'rolodex_jumbo_pencil_cup',
'champion_copper_plus_spark_plug',
'oreo_mega_stuf',
]
objects = jsk_apc2015_common.get_object_list()
target_obj = None
while (target_obj is None) or (target_obj in abandon_objects):
candidates = []
for i in xrange(N):
i_obj = np.random.randint(0, len(objects))
candidates.append(objects[i_obj])
i_target = np.random.randint(0, len(candidates))
target_obj = candidates[i_target]
json_data = {
'bin_contents': {
'bin_{0}'.format(target_bin.upper()): candidates,
},
'work_order': [
def __init__(self):
self.file_name = 'rgb'
self.object_names = jsk_apc2015_common.get_object_list()
self.cfeatures = []
self.labels = []
