def test_task():
dataset = Task.Dataset.from_voc('https://autogluon.s3.amazonaws.com/datasets/tiny_motorbike.zip')
train_data, _, test_data = dataset.random_split()
detector = Task()
detector.fit(train_data, num_trials=1, hyperparameters={'batch_size': 4, 'epochs': 5, 'early_stop_max_value': 0.2})
test_result = detector.predict(test_data)
print('test result', test_result)
detector.save('detector.ag')
detector2 = Task.load('detector.ag')
fit_summary = detector2.fit_summary()
test_map = detector2.evaluate(test_data)
test_result2 = detector2.predict(test_data)
assert test_result2.equals(test_result)
def test_task():
dataset = Task.Dataset.from_voc('https://autogluon.s3.amazonaws.com/datasets/tiny_motorbike.zip')
train_data, _, test_data = dataset.random_split()
detector = Task()
detector.fit(train_data, hyperparameters={'batch_size': 4, 'epochs': 5, 'early_stop_max_value': 0.2}, hyperparameter_tune_kwargs={'num_trials': 1})
test_result = detector.predict(test_data)
detector.save('detector.ag')
detector2 = Task.load('detector.ag')
fit_summary = detector2.fit_summary()
test_map = detector2.evaluate(test_data)
test_result2 = detector2.predict(test_data)
assert test_result2.equals(test_result), f'{test_result2} != \n {test_result}'
# to numpy
test_result2 = detector2.predict(test_data, as_pandas=False)
import numpy as np
from matplotlib import pyplot as plt
from mxnet import image
from gluoncv import utils
from autogluon.vision import ObjectDetector
detector = ObjectDetector.load('enemy_detector.ag')
image_array = image.imread('test.jpg')
result = detector.predict(image_array)
selected_result = result.query('predict_score > 0.85')
class_ids , class_names = selected_result['predict_class'].factorize()
bounding_boxes = np.array([[x[i] for i in x.keys()] for x in list(selected_result['predict_rois'])])
scores = np.array(selected_result['predict_score'])
utils.viz.plot_bbox(image_array, bounding_boxes, scores=scores,
labels=class_ids, class_names = class_names, absolute_coordinates=False)
plt.savefig('result.jpg')
config = yaml.safe_load(f) # AutoGluon-specific config
if args.n_gpus:
config['num_gpus'] = int(args.n_gpus)
# ---------------------------------------------------------------- Training
train_dataset = ObjectDetector.Dataset.from_voc(config['dataset'],
splits='trainval')
ag_predictor_args = config['ag_predictor_args']
ag_predictor_args['path'] = args.model_dir
ag_fit_args = config['ag_fit_args']
print('Running training job with the config:')
pprint(config)
predictor = ObjectDetector(**ag_predictor_args).fit(
train_dataset, **ag_fit_args)
predictor.save(f'{args.model_dir}/predictor.pkl')
# --------------------------------------------------------------- Inference
test_dataset = ObjectDetector.Dataset.from_voc(config['dataset'],
splits='test')
y_pred = predictor.predict(test_dataset)
if config.get('output_prediction_format', 'csv') == 'parquet':
y_pred.to_parquet(f'{args.output_data_dir}/predictions.parquet')
else:
y_pred.to_csv(f'{args.output_data_dir}/predictions.csv')
[30, 35], [20]),
'warmup_iters':
ag.Int(5, 500),
'wd':
ag.Categorical(1e-4, 5e-4, 2.5e-4),
'syncbn':
True,
'label_smooth':
False,
'epochs':
ag.Categorical(30, 40, 50, 60),
'transfer':
transfer
}
kwargs = {
'num_trials': args.num_trials,
'nthreads_per_trial': 16,
'ngpus_per_trial': 8,
'time_limit': time_limit,
'dist_ip_addrs': [],
'hyperparameters': hyperparameters
}
else:
raise NotImplementedError('%s is not implemented.', args.meta_arch)
detector = ObjectDetector()
detector.fit(dataset_train, **kwargs)
test_map = detector.evaluate(dataset_test)
print("mAP on test dataset: {}".format(test_map[-1][-1]))
print(test_map)
detector.save('final_model.model')
from autogluon.vision import ObjectDetector
dataset_train = ObjectDetector.Dataset.from_voc('train_data', splits='train')
time_limit = 5 * 60 * 60 # 5 hour
detector = ObjectDetector()
hyperparameters = {'batch_size': 4, 'epochs': 2}
detector.fit(dataset_train,
time_limit=time_limit,
hyperparameters=hyperparameters)
detector.save('enemy_detector.ag')