def __init__(self,filename_weights,path_root,obj_threshold=0.5):
config = InferenceConfig()
self.model = detector_Mask_RCNN_core.MaskRCNN(mode="inference", model_dir=path_root, config=config)
self.model.load_weights(filename_weights, by_name=True)
self.class_names = ['BG']+tools_YOLO.get_COCO_class_names()
self.colors = tools_YOLO.generate_colors(len(self.class_names) - 1)
self.colors.insert(0, self.colors[-1])
self.obj_threshold = obj_threshold
return
def plot_mAP_overlap(folder_annotation,
file_markup_true,
file_markup_pred,
filename_meta,
folder_out,
out_prefix='',
delim=' '):
input_image_size, class_names, anchors, anchor_mask, obj_threshold, nms_threshold = tools_YOLO.load_metadata(
filename_meta)
colors = tools_YOLO.generate_colors(len(class_names))
ovp_ths = [0.9, 0.8, 0.7, 0.6, 0.5, 0.1]
ovd_ths = [0.1, 0.2, 0.5, 0.99]
results = []
for ovp_th in ovp_ths:
out_dir = folder_out + 'ovp_%02d/' % int(ovp_th * 100)
if not os.path.exists(out_dir):
os.makedirs(out_dir)
for ovd_th in ovd_ths:
precisions, recalls, confidences, class_IDs = get_precsion_recall_data_from_markups(
folder_annotation,
file_markup_true,
file_markup_pred,
None,
ovp_th,
ovd_th,
delim=delim)
mAP = 0
for i, class_ID in enumerate(class_IDs):
if len(precisions[i]) > 1:
xxx = numpy.array(
[precisions[i], recalls[i], confidences[i]]).T
filename_out = out_dir + out_prefix + 'OVD_%02d_AP_%02d_%s.txt' % (
int(ovd_th * 100), class_ID, class_names[class_ID])
tools_IO.save_mat(xxx, filename_out)
filename_out = out_dir + out_prefix + 'OVD_%02d_AP_%02d_%s.png' % (
int(ovd_th * 100), class_ID, class_names[class_ID])
AP = write_precision_recall(
filename_out,
precisions[i],
recalls[i],
caption='ovp %1.2f ovd %1.2f ' % (ovp_th, ovd_th) +
class_names[class_ID],
color=(colors[class_ID][2] / 255.0,
colors[class_ID][1] / 255.0,
colors[class_ID][0] / 255.0))
mAP += AP
results.append(mAP / len(class_IDs))
return results[0]
def __init__(self, filename_topology, obj_threshold=0.5):
self.obj_threshold = obj_threshold
self.class_names = ['BG'] + tools_YOLO.get_COCO_class_names()
self.colors = tools_YOLO.generate_colors(len(self.class_names) - 1)
self.colors.insert(0, self.colors[-1])
with tf.gfile.FastGFile(filename_topology, 'rb') as f:
self.graph_def = tf.GraphDef()
self.graph_def.ParseFromString(f.read())
return
def plot_mAP_iou(folder_annotation,
file_markup_true,
file_markup_pred,
filename_meta,
folder_out,
out_prefix='',
delim=' '):
input_image_size, class_names, anchors, anchor_mask, obj_threshold, nms_threshold = tools_YOLO.load_metadata(
filename_meta)
colors = tools_YOLO.generate_colors(len(class_names))
iuo_ths = [0.5, 0.3, 0.1, 0.01]
results = []
for iuo_th in iuo_ths:
ovp_th, ovd_th = None, None
out_dir = folder_out + 'iou_%02d/' % int(iuo_th * 100)
if not os.path.exists(out_dir):
os.makedirs(out_dir)
precisions, recalls, confidences, class_IDs = get_precsion_recall_data_from_markups(
folder_annotation,
file_markup_true,
file_markup_pred,
iuo_th,
ovp_th,
ovd_th,
delim=delim)
mAP = 0
for i, class_ID in enumerate(class_IDs):
if len(precisions[i]) > 1:
filename_out = out_dir + out_prefix + 'AP_%02d_%s.png' % (
class_ID, class_names[class_ID])
AP = write_precision_recall(
filename_out,
precisions[i],
recalls[i],
caption='iou %1.2f ' % iuo_th + class_names[class_ID],
color=(colors[class_ID][2] / 255.0,
colors[class_ID][1] / 255.0,
colors[class_ID][0] / 255.0))
mAP += AP
results.append(mAP / len(class_IDs))
print(out_prefix, iuo_th, mAP / len(class_IDs))
print()
return results[0]
def init_tiny(self, default_weights_h5, num_classes):
self.input_image_size, self.class_names, self.anchors, self.anchor_mask, self.obj_threshold, self.nms_threshold = tools_YOLO.init_default_metadata_tiny(
num_classes)
self.colors = tools_YOLO.generate_colors(len(self.class_names))
self.input_tensor_shape = K.placeholder(shape=(2, ))
self.sess = K.get_session()
self.model = detector_YOLO3_core.yolo_body_tiny(
Input(shape=(None, None, 3)), 3, num_classes)
default_model = load_model(default_weights_h5, compile=False)
detector_YOLO3_core.assign_weights(default_model, self.model)
self.boxes, self.scores, self.classes = detector_YOLO3_core.get_tensors_box_score_class(
self.model.output,
self.anchors,
self.anchor_mask,
len(self.class_names),
self.input_tensor_shape,
score_threshold=self.obj_threshold,
iou_threshold=self.nms_threshold)
return
def __load_model(self,
model_weights_h5,
filename_metadata,
obj_threshold=None):
self.input_image_size, self.class_names, self.anchors, self.anchor_mask, self.obj_threshold, self.nms_threshold = tools_YOLO.load_metadata(
filename_metadata)
if obj_threshold is not None: self.obj_threshold = obj_threshold
self.colors = tools_YOLO.generate_colors(len(self.class_names))
self.input_tensor_shape = K.placeholder(shape=(2, ))
self.sess = K.get_session()
self.model = load_model(model_weights_h5, compile=False)
self.boxes, self.scores, self.classes = detector_YOLO3_core.get_tensors_box_score_class(
self.model.output,
self.anchors,
self.anchor_mask,
len(self.class_names),
self.input_tensor_shape,
score_threshold=self.obj_threshold,
iou_threshold=self.nms_threshold)
return
def generate_colors(N):
return tools_YOLO.generate_colors(N)