def __init__(self, mode=None, config_path=None, *args, **kwargs):
if config_path is None:
config_path = utils.Path(inspect.getfile(
self.__class__)).with_name('config.yaml')
if mode is None:
self.processor = Processor()
elif config_path.exists():
config = utils.load_yaml(config_path)
self.processor = utils.create_instance(config[mode], *args,
**kwargs)
else:
raise FileNotFoundError('{} not found.'.format(config_path))
def on_notepad(self, filepath):
utils.spawnProcess('notepad "%s"' % utils.Path(filepath).asNative())
import utils
# Filter future deprecation warnings
utils.filterwarnings('ignore')
root = utils.Path("../ip102_v1.1")
model_dir = utils.Path("models/pest")
detector = utils.PestDetector(root=root, model_dir=model_dir)
# Preparing data and model with an optional quantization
detector.getdata(bs=32, num_workers=1, noise=False, blur=False, basic=True)
detector.createmodel(quantize=True)
# Training model
detector.train(epochs=10, firstrun=True, min_lr=None, interpret=False)
detector.train(epochs=20,
firstrun=False,
min_lr=utils.asklr(),
interpret=False)
# Pruning model and finetuning further
detector.loadmodel(path=model_dir / "recentbest")
detector.prunemodel(amount=0.5)
detector.findlr()
detector.train(epochs=20, firstrun=False, min_lr=utils.asklr(), interpret=True)
# Finish quantization
detector.quantize()
# Save model for mobile deployment using torchscript
def initialize_paths(game):
paths = []
for path in utils.possible_paths:
paths.append(utils.Path(path, game))
return paths