def get_YOLOv2_Predictor():
# has been trained with coco
file_yolov2_cfg = getDataFile("yolov2/yolov2.cfg")
yolov2_training_ctx = TrainingContext(
n_classes = 80,
trainfile = "/home/pjreddie/data/coco/trainvalno5k.txt",
validfile = "coco_testdev",
namefile = getDataFile("yolov3/coco.names"), # all the other values don't make any difference when predicting
backup_dir ="/home/pjreddie/backup/",
setname = "coco"
)
return Predictor(training_ctx = yolov2_training_ctx, weight_file = get_yolov2_weights_file(), config_file = file_yolov2_cfg, thresh = 0.6)
def __init__(self,
training_ctx=None,
config_file=None,
weight_file=None,
thresh=0.9,
hier_thresh=0.5,
verbose=False,
gpu_index=-1):
# datacfg(datacfg), cfgfile(cfgfile), weightfile(weightfile), datadir(datadir), thresh(thresh), hier_thresh(hier_thresh)
"""
DarknetContext ctx = DarknetContext(
darkdir + std::string("cfg/coco.data"), # how the weights were trained?
darkdir + std::string("cfg/yolov3.cfg"), # convoluted neural network scheme
darkdir + std::string("yolov3.weights"), # neural network weights. has been trained with the "coco" data
darkdir + std::string("data/labels/"), # alphabet bitmaps
0.1, 0.5);
"""
self.pre = pre_mod + self.__class__.__name__ + " : "
assert (isinstance(training_ctx, TrainingContext))
assert (isinstance(config_file, str))
assert (isinstance(weight_file, str))
assert (isinstance(thresh, float))
assert (isinstance(hier_thresh, float))
assert (os.path.exists(config_file))
assert (os.path.exists(weight_file))
self.verbose = verbose
if (self.verbose):
print(self.pre)
print(self.pre, str(training_ctx))
print(self.pre)
print(self.pre, config_file)
print(self.pre, weight_file)
print(self.pre, getDataFile("labels/"))
self.thresh = thresh
self.hier_thresh = hier_thresh
# return
self.ctx = core.DarknetContext(str(training_ctx), config_file,
weight_file, getDataFile("labels/"))
self.predictor = core.DarknetPredictor(self.ctx, thresh, hier_thresh)
if (gpu_index >= 0):
self.predictor.setGpuIndex(gpu_index)