Object Detection - Quick Start¶

Object detection is the process of identifying and localizing objects in an image and is an important task in computer vision. Follow this tutorial to learn how to use AutoGluon for object detection.

Tip: If you are new to AutoGluon, review Image Classification - Quick Start first to learn the basics of the AutoGluon API.

Our goal is to detect motorbike in images by YOLOv3 model. A tiny dataset is collected from VOC dataset, which only contains the motorbike category. The model pretrained on the COCO dataset is used to fine-tune our small dataset. With the help of AutoGluon, we are able to try many models with different hyperparameters automatically, and return the best one as our final model.

To start, import autogluon.vision and ObjectDetection module as your task:

import autogluon.core as ag
from autogluon.vision import ObjectDetection as task

Tiny_motorbike Dataset¶

We collect a toy dataset for detecting motorbikes in images. From the VOC dataset, images are randomly selected for training, validation, and testing - 120 images for training, 50 images for validation, and 50 for testing. This tiny dataset follows the same format as VOC.

Using the commands below, we can download this dataset, which is only 23M. The variable root specifies the path to store the dataset in. The name of unzipped folder is called tiny_motorbike.

root = './'
filename_zip = ag.download('https://autogluon.s3.amazonaws.com/datasets/tiny_motorbike.zip',
                        path=root)
filename = ag.unzip(filename_zip, root=root)

21273KB [00:00, 64530.27KB/s]

When we retrieve the dataset, we can create a dataset instance with its path and classes if it is a custom dataset.

import os
data_root = os.path.join(root, filename)
dataset_train = task.Dataset(data_root, classes=('motorbike',))

Fit Models by AutoGluon¶

In this section, we demonstrate how to apply AutoGluon to fit our detection models. We use mobilenet as the backbone for the YOLOv3 model. Two different learning rates are used to fine-tune the network. The best model is the one that obtains the best performance on the validation dataset. You can also try using more networks and hyperparameters to create a larger searching space.

We fit a classifier using AutoGluon as follows. In each experiment (one trial in our searching space), we train the model for 30 epochs.

time_limits = 5*60*60  # 5 hours
epochs = 30
detector = task.fit(dataset_train,
                    num_trials=2,
                    epochs=epochs,
                    lr=ag.Categorical(5e-4, 1e-4),
                    ngpus_per_trial=1,
                    time_limits=time_limits)

scheduler: FIFOScheduler(
DistributedResourceManager{
(Remote: Remote REMOTE_ID: 0,
    <Remote: 'inproc://172.31.35.213/10487/1' processes=1 threads=8, memory=33.24 GB>, Resource: NodeResourceManager(8 CPUs, 1 GPUs))
})

HBox(children=(HTML(value=''), FloatProgress(value=0.0, max=2.0), HTML(value='')))

INFO:root:{'meta_arch': 'yolo3', 'dataset': <autogluon.vision.object_detection.dataset.voc.CustomVOCDetection object at 0x7f610a5cbad0>, 'net': 'mobilenet1.0', 'lr': 0.0005, 'loss': SoftmaxCrossEntropyLoss(batch_axis=0, w=None), 'num_gpus': 1, 'batch_size': 16, 'split_ratio': 0.8, 'epochs': 30, 'num_workers': 8, 'hybridize': True, 'verbose': False, 'final_fit': False, 'seed': 223, 'data_shape': 416, 'start_epoch': 0, 'transfer': 'coco', 'lr_mode': 'step', 'lr_decay': 0.1, 'lr_decay_period': 0, 'lr_decay_epoch': '160,180', 'warmup_lr': 0.0, 'warmup_epochs': 2, 'warmup_iters': 1000, 'warmup_factor': 0.3333333333333333, 'momentum': 0.9, 'wd': 0.0005, 'log_interval': 100, 'save_prefix': 'yolo3_mobilenet1.0_custom', 'save_interval': 10, 'val_interval': 1, 'num_samples': -1, 'no_random_shape': False, 'no_wd': False, 'mixup': False, 'no_mixup_epochs': 20, 'label_smooth': False, 'resume': '', 'syncbn': False, 'reuse_pred_weights': True, 'task_id': 0}
INFO:root:[Epoch 0] Training cost: 5.550, ObjLoss=743.047,BoxCenterLoss=3.541,BoxScaleLoss=2.480,ClassLoss=0.916
INFO:root:[Epoch 0] Validation: motorbike=0.0 mAP=0.0
INFO:root:[Epoch 1] Training cost: 4.211, ObjLoss=33.338,BoxCenterLoss=3.683,BoxScaleLoss=1.624,ClassLoss=0.835
INFO:root:[Epoch 1] Validation: motorbike=0.0303030303030303 mAP=0.0303030303030303
INFO:root:[Epoch 2] Training cost: 6.221, ObjLoss=21.797,BoxCenterLoss=3.657,BoxScaleLoss=1.409,ClassLoss=0.717
INFO:root:[Epoch 2] Validation: motorbike=0.15229127960619673 mAP=0.15229127960619673
INFO:root:[Epoch 3] Training cost: 3.538, ObjLoss=20.184,BoxCenterLoss=3.721,BoxScaleLoss=1.396,ClassLoss=0.694
INFO:root:[Epoch 3] Validation: motorbike=0.18229599979355463 mAP=0.18229599979355463
INFO:root:[Epoch 4] Training cost: 3.730, ObjLoss=17.578,BoxCenterLoss=3.625,BoxScaleLoss=1.335,ClassLoss=0.606
INFO:root:[Epoch 4] Validation: motorbike=0.4485862052408224 mAP=0.4485862052408224
INFO:root:[Epoch 5] Training cost: 3.703, ObjLoss=14.644,BoxCenterLoss=3.744,BoxScaleLoss=1.205,ClassLoss=0.646
INFO:root:[Epoch 5] Validation: motorbike=0.35256400485131256 mAP=0.35256400485131256
INFO:root:[Epoch 6] Training cost: 6.271, ObjLoss=13.078,BoxCenterLoss=3.590,BoxScaleLoss=1.235,ClassLoss=0.498
INFO:root:[Epoch 6] Validation: motorbike=0.3853689638378634 mAP=0.3853689638378634
INFO:root:[Epoch 7] Training cost: 4.607, ObjLoss=14.107,BoxCenterLoss=3.406,BoxScaleLoss=1.114,ClassLoss=0.452
INFO:root:[Epoch 7] Validation: motorbike=0.5105032902302911 mAP=0.5105032902302911
INFO:root:[Epoch 8] Training cost: 6.793, ObjLoss=10.122,BoxCenterLoss=3.493,BoxScaleLoss=1.056,ClassLoss=0.321
INFO:root:[Epoch 8] Validation: motorbike=0.6566899150232484 mAP=0.6566899150232484
INFO:root:[Epoch 9] Training cost: 6.640, ObjLoss=10.550,BoxCenterLoss=3.332,BoxScaleLoss=0.899,ClassLoss=0.312
INFO:root:[Epoch 9] Validation: motorbike=0.5967563113264563 mAP=0.5967563113264563
INFO:root:[Epoch 10] Training cost: 6.787, ObjLoss=8.834,BoxCenterLoss=3.437,BoxScaleLoss=0.944,ClassLoss=0.277
INFO:root:[Epoch 10] Validation: motorbike=0.6796814296814296 mAP=0.6796814296814296
INFO:root:[Epoch 11] Training cost: 3.996, ObjLoss=12.229,BoxCenterLoss=3.472,BoxScaleLoss=1.033,ClassLoss=0.368
INFO:root:[Epoch 11] Validation: motorbike=0.7030695814005539 mAP=0.7030695814005539
INFO:root:[Epoch 12] Training cost: 5.385, ObjLoss=6.838,BoxCenterLoss=3.458,BoxScaleLoss=0.854,ClassLoss=0.275
INFO:root:[Epoch 12] Validation: motorbike=0.6964525119209569 mAP=0.6964525119209569
INFO:root:[Epoch 13] Training cost: 3.461, ObjLoss=9.435,BoxCenterLoss=3.334,BoxScaleLoss=0.964,ClassLoss=0.362
INFO:root:[Epoch 13] Validation: motorbike=0.5928087324861518 mAP=0.5928087324861518
INFO:root:[Epoch 14] Training cost: 3.659, ObjLoss=9.404,BoxCenterLoss=3.517,BoxScaleLoss=0.953,ClassLoss=0.310
INFO:root:[Epoch 14] Validation: motorbike=0.685586711484716 mAP=0.685586711484716
INFO:root:[Epoch 15] Training cost: 4.722, ObjLoss=8.980,BoxCenterLoss=3.307,BoxScaleLoss=0.811,ClassLoss=0.321
INFO:root:[Epoch 15] Validation: motorbike=0.7307586316122902 mAP=0.7307586316122902
INFO:root:[Epoch 16] Training cost: 4.014, ObjLoss=7.042,BoxCenterLoss=3.343,BoxScaleLoss=0.853,ClassLoss=0.264
INFO:root:[Epoch 16] Validation: motorbike=0.6871621621621622 mAP=0.6871621621621622
INFO:root:[Epoch 17] Training cost: 3.963, ObjLoss=5.720,BoxCenterLoss=3.283,BoxScaleLoss=0.856,ClassLoss=0.265
INFO:root:[Epoch 17] Validation: motorbike=0.7054391987793628 mAP=0.7054391987793628
INFO:root:[Epoch 18] Training cost: 3.491, ObjLoss=5.835,BoxCenterLoss=3.557,BoxScaleLoss=0.921,ClassLoss=0.266
INFO:root:[Epoch 18] Validation: motorbike=0.7255157708439132 mAP=0.7255157708439132
INFO:root:[Epoch 19] Training cost: 3.419, ObjLoss=6.228,BoxCenterLoss=3.513,BoxScaleLoss=0.877,ClassLoss=0.289
INFO:root:[Epoch 19] Validation: motorbike=0.7673262466783052 mAP=0.7673262466783052
INFO:root:[Epoch 20] Training cost: 3.772, ObjLoss=5.422,BoxCenterLoss=3.262,BoxScaleLoss=0.828,ClassLoss=0.228
INFO:root:[Epoch 20] Validation: motorbike=0.7142471821264391 mAP=0.7142471821264391
INFO:root:[Epoch 21] Training cost: 6.505, ObjLoss=5.785,BoxCenterLoss=3.386,BoxScaleLoss=0.909,ClassLoss=0.150
INFO:root:[Epoch 21] Validation: motorbike=0.797701964859583 mAP=0.797701964859583
INFO:root:[Epoch 22] Training cost: 5.254, ObjLoss=5.973,BoxCenterLoss=3.554,BoxScaleLoss=0.896,ClassLoss=0.183
INFO:root:[Epoch 22] Validation: motorbike=0.7634377602665129 mAP=0.7634377602665129
INFO:root:[Epoch 23] Training cost: 3.650, ObjLoss=5.175,BoxCenterLoss=3.372,BoxScaleLoss=0.832,ClassLoss=0.195
INFO:root:[Epoch 23] Validation: motorbike=0.724888125573057 mAP=0.724888125573057
INFO:root:[Epoch 24] Training cost: 4.904, ObjLoss=5.621,BoxCenterLoss=3.623,BoxScaleLoss=0.948,ClassLoss=0.200
INFO:root:[Epoch 24] Validation: motorbike=0.7301328939259975 mAP=0.7301328939259975
INFO:root:[Epoch 25] Training cost: 4.366, ObjLoss=5.174,BoxCenterLoss=3.317,BoxScaleLoss=0.819,ClassLoss=0.195
INFO:root:[Epoch 25] Validation: motorbike=0.8037408433876359 mAP=0.8037408433876359
INFO:root:[Epoch 26] Training cost: 4.785, ObjLoss=4.920,BoxCenterLoss=3.256,BoxScaleLoss=0.842,ClassLoss=0.153
INFO:root:[Epoch 26] Validation: motorbike=0.7520477415214258 mAP=0.7520477415214258
INFO:root:[Epoch 27] Training cost: 5.787, ObjLoss=5.384,BoxCenterLoss=3.339,BoxScaleLoss=0.717,ClassLoss=0.132
INFO:root:[Epoch 27] Validation: motorbike=0.8062521877835017 mAP=0.8062521877835017
INFO:root:[Epoch 28] Training cost: 4.132, ObjLoss=4.931,BoxCenterLoss=3.181,BoxScaleLoss=0.758,ClassLoss=0.128
INFO:root:[Epoch 28] Validation: motorbike=0.8308813566260375 mAP=0.8308813566260375
INFO:root:[Epoch 29] Training cost: 5.582, ObjLoss=4.901,BoxCenterLoss=3.319,BoxScaleLoss=0.827,ClassLoss=0.112
INFO:root:[Epoch 29] Validation: motorbike=0.752907541779015 mAP=0.752907541779015

INFO:root:{'meta_arch': 'yolo3', 'dataset': <autogluon.vision.object_detection.dataset.voc.CustomVOCDetection object at 0x7f610a5cb650>, 'net': 'mobilenet1.0', 'lr': 0.0001, 'loss': SoftmaxCrossEntropyLoss(batch_axis=0, w=None), 'num_gpus': 1, 'batch_size': 16, 'split_ratio': 0.8, 'epochs': 30, 'num_workers': 8, 'hybridize': True, 'verbose': False, 'final_fit': False, 'seed': 223, 'data_shape': 416, 'start_epoch': 0, 'transfer': 'coco', 'lr_mode': 'step', 'lr_decay': 0.1, 'lr_decay_period': 0, 'lr_decay_epoch': '160,180', 'warmup_lr': 0.0, 'warmup_epochs': 2, 'warmup_iters': 1000, 'warmup_factor': 0.3333333333333333, 'momentum': 0.9, 'wd': 0.0005, 'log_interval': 100, 'save_prefix': 'yolo3_mobilenet1.0_custom', 'save_interval': 10, 'val_interval': 1, 'num_samples': -1, 'no_random_shape': False, 'no_wd': False, 'mixup': False, 'no_mixup_epochs': 20, 'label_smooth': False, 'resume': '', 'syncbn': False, 'reuse_pred_weights': True, 'task_id': 1}
INFO:root:[Epoch 0] Training cost: 5.788, ObjLoss=887.231,BoxCenterLoss=3.610,BoxScaleLoss=2.740,ClassLoss=1.016
INFO:root:[Epoch 0] Validation: motorbike=0.012692634262756212 mAP=0.012692634262756212
INFO:root:[Epoch 1] Training cost: 4.289, ObjLoss=14.288,BoxCenterLoss=3.433,BoxScaleLoss=2.089,ClassLoss=0.914
INFO:root:[Epoch 1] Validation: motorbike=0.0 mAP=0.0
INFO:root:[Epoch 2] Training cost: 6.410, ObjLoss=14.401,BoxCenterLoss=3.805,BoxScaleLoss=1.889,ClassLoss=0.838
INFO:root:[Epoch 2] Validation: motorbike=0.15584415584415584 mAP=0.15584415584415584
INFO:root:[Epoch 3] Training cost: 3.862, ObjLoss=15.998,BoxCenterLoss=3.835,BoxScaleLoss=1.797,ClassLoss=0.769
INFO:root:[Epoch 3] Validation: motorbike=0.3375879165225412 mAP=0.3375879165225412
INFO:root:[Epoch 4] Training cost: 3.913, ObjLoss=9.353,BoxCenterLoss=3.757,BoxScaleLoss=1.570,ClassLoss=0.707
INFO:root:[Epoch 4] Validation: motorbike=0.3652805794975243 mAP=0.3652805794975243
INFO:root:[Epoch 5] Training cost: 3.575, ObjLoss=8.162,BoxCenterLoss=3.613,BoxScaleLoss=1.372,ClassLoss=0.713
INFO:root:[Epoch 5] Validation: motorbike=0.44498872046886784 mAP=0.44498872046886784
INFO:root:[Epoch 6] Training cost: 6.561, ObjLoss=7.205,BoxCenterLoss=3.814,BoxScaleLoss=1.249,ClassLoss=0.587
INFO:root:[Epoch 6] Validation: motorbike=0.4495434154924295 mAP=0.4495434154924295
INFO:root:[Epoch 7] Training cost: 4.627, ObjLoss=6.068,BoxCenterLoss=3.469,BoxScaleLoss=1.187,ClassLoss=0.553
INFO:root:[Epoch 7] Validation: motorbike=0.6412523856072242 mAP=0.6412523856072242
INFO:root:[Epoch 8] Training cost: 7.046, ObjLoss=6.776,BoxCenterLoss=3.643,BoxScaleLoss=1.114,ClassLoss=0.512
INFO:root:[Epoch 8] Validation: motorbike=0.6233064962036925 mAP=0.6233064962036925
INFO:root:[Epoch 9] Training cost: 6.995, ObjLoss=6.000,BoxCenterLoss=3.550,BoxScaleLoss=1.110,ClassLoss=0.447
INFO:root:[Epoch 9] Validation: motorbike=0.5812085703846757 mAP=0.5812085703846757
INFO:root:[Epoch 10] Training cost: 6.900, ObjLoss=5.855,BoxCenterLoss=3.344,BoxScaleLoss=0.926,ClassLoss=0.391
INFO:root:[Epoch 10] Validation: motorbike=0.652847267441619 mAP=0.652847267441619
INFO:root:[Epoch 11] Training cost: 4.017, ObjLoss=6.185,BoxCenterLoss=3.682,BoxScaleLoss=1.180,ClassLoss=0.535
INFO:root:[Epoch 11] Validation: motorbike=0.6612478073223106 mAP=0.6612478073223106
INFO:root:[Epoch 12] Training cost: 5.256, ObjLoss=5.306,BoxCenterLoss=3.348,BoxScaleLoss=1.014,ClassLoss=0.368
INFO:root:[Epoch 12] Validation: motorbike=0.7191117400479203 mAP=0.7191117400479203
INFO:root:[Epoch 13] Training cost: 3.227, ObjLoss=5.569,BoxCenterLoss=3.519,BoxScaleLoss=1.106,ClassLoss=0.534
INFO:root:[Epoch 13] Validation: motorbike=0.6957360820997184 mAP=0.6957360820997184
INFO:root:[Epoch 14] Training cost: 4.036, ObjLoss=5.235,BoxCenterLoss=3.521,BoxScaleLoss=0.994,ClassLoss=0.413
INFO:root:[Epoch 14] Validation: motorbike=0.7275157275157275 mAP=0.7275157275157275
INFO:root:[Epoch 15] Training cost: 4.545, ObjLoss=4.926,BoxCenterLoss=3.411,BoxScaleLoss=0.957,ClassLoss=0.410
INFO:root:[Epoch 15] Validation: motorbike=0.7640561458743278 mAP=0.7640561458743278
INFO:root:[Epoch 16] Training cost: 4.503, ObjLoss=5.069,BoxCenterLoss=3.626,BoxScaleLoss=0.979,ClassLoss=0.404
INFO:root:[Epoch 16] Validation: motorbike=0.7610060394151305 mAP=0.7610060394151305
INFO:root:[Epoch 17] Training cost: 3.892, ObjLoss=4.888,BoxCenterLoss=3.515,BoxScaleLoss=1.061,ClassLoss=0.391
INFO:root:[Epoch 17] Validation: motorbike=0.7631065903793176 mAP=0.7631065903793176
INFO:root:[Epoch 18] Training cost: 3.495, ObjLoss=4.825,BoxCenterLoss=3.517,BoxScaleLoss=0.984,ClassLoss=0.365
INFO:root:[Epoch 18] Validation: motorbike=0.7979947759065183 mAP=0.7979947759065183
INFO:root:[Epoch 19] Training cost: 3.564, ObjLoss=4.675,BoxCenterLoss=3.507,BoxScaleLoss=1.033,ClassLoss=0.380
INFO:root:[Epoch 19] Validation: motorbike=0.7857805624254514 mAP=0.7857805624254514
INFO:root:[Epoch 20] Training cost: 3.550, ObjLoss=4.226,BoxCenterLoss=3.223,BoxScaleLoss=0.847,ClassLoss=0.315
INFO:root:[Epoch 20] Validation: motorbike=0.8069452792901995 mAP=0.8069452792901995
INFO:root:[Epoch 21] Training cost: 6.567, ObjLoss=5.112,BoxCenterLoss=3.406,BoxScaleLoss=0.909,ClassLoss=0.272
INFO:root:[Epoch 21] Validation: motorbike=0.7986070282721 mAP=0.7986070282721
INFO:root:[Epoch 22] Training cost: 5.198, ObjLoss=4.847,BoxCenterLoss=3.548,BoxScaleLoss=0.964,ClassLoss=0.278
INFO:root:[Epoch 22] Validation: motorbike=0.8159504102717108 mAP=0.8159504102717108
INFO:root:[Epoch 23] Training cost: 3.508, ObjLoss=4.396,BoxCenterLoss=3.299,BoxScaleLoss=0.932,ClassLoss=0.324
INFO:root:[Epoch 23] Validation: motorbike=0.8112025528815524 mAP=0.8112025528815524
INFO:root:[Epoch 24] Training cost: 4.728, ObjLoss=4.640,BoxCenterLoss=3.524,BoxScaleLoss=0.919,ClassLoss=0.280
INFO:root:[Epoch 24] Validation: motorbike=0.8241794439151704 mAP=0.8241794439151704
INFO:root:[Epoch 25] Training cost: 4.696, ObjLoss=4.223,BoxCenterLoss=3.474,BoxScaleLoss=0.953,ClassLoss=0.262
INFO:root:[Epoch 25] Validation: motorbike=0.8242634262639366 mAP=0.8242634262639366
INFO:root:[Epoch 26] Training cost: 4.629, ObjLoss=4.398,BoxCenterLoss=3.582,BoxScaleLoss=0.852,ClassLoss=0.270
INFO:root:[Epoch 26] Validation: motorbike=0.8561498310302138 mAP=0.8561498310302138
INFO:root:[Epoch 27] Training cost: 6.108, ObjLoss=4.261,BoxCenterLoss=3.357,BoxScaleLoss=0.905,ClassLoss=0.215
INFO:root:[Epoch 27] Validation: motorbike=0.8323144130766638 mAP=0.8323144130766638
INFO:root:[Epoch 28] Training cost: 4.251, ObjLoss=4.186,BoxCenterLoss=3.146,BoxScaleLoss=0.778,ClassLoss=0.215
INFO:root:[Epoch 28] Validation: motorbike=0.828824126431782 mAP=0.828824126431782
INFO:root:[Epoch 29] Training cost: 5.650, ObjLoss=4.612,BoxCenterLoss=3.527,BoxScaleLoss=0.901,ClassLoss=0.238
INFO:root:[Epoch 29] Validation: motorbike=0.8320460895292883 mAP=0.8320460895292883
INFO:root:{'meta_arch': 'yolo3', 'dataset': <autogluon.vision.object_detection.dataset.voc.CustomVOCDetection object at 0x7f610a5cb4d0>, 'net': 'mobilenet1.0', 'lr': 0.0001, 'loss': SoftmaxCrossEntropyLoss(batch_axis=0, w=None), 'num_gpus': 1, 'batch_size': 16, 'split_ratio': 0.8, 'epochs': 30, 'num_workers': 8, 'hybridize': True, 'verbose': False, 'final_fit': True, 'seed': 223, 'data_shape': 416, 'start_epoch': 0, 'transfer': 'coco', 'lr_mode': 'step', 'lr_decay': 0.1, 'lr_decay_period': 0, 'lr_decay_epoch': '160,180', 'warmup_lr': 0.0, 'warmup_epochs': 2, 'warmup_iters': 1000, 'warmup_factor': 0.3333333333333333, 'momentum': 0.9, 'wd': 0.0005, 'log_interval': 100, 'save_prefix': 'yolo3_mobilenet1.0_custom', 'save_interval': 10, 'val_interval': 1, 'num_samples': -1, 'no_random_shape': False, 'no_wd': False, 'mixup': False, 'no_mixup_epochs': 20, 'label_smooth': False, 'resume': '', 'syncbn': False, 'reuse_pred_weights': True, 'task_id': 2}
INFO:root:[Epoch 0] Training cost: 6.367, ObjLoss=713.653,BoxCenterLoss=3.844,BoxScaleLoss=2.822,ClassLoss=1.006
INFO:root:[Epoch 1] Training cost: 5.245, ObjLoss=15.767,BoxCenterLoss=3.840,BoxScaleLoss=2.086,ClassLoss=0.910
INFO:root:[Epoch 2] Training cost: 6.454, ObjLoss=13.244,BoxCenterLoss=3.711,BoxScaleLoss=1.675,ClassLoss=0.746
INFO:root:[Epoch 3] Training cost: 9.119, ObjLoss=9.872,BoxCenterLoss=3.644,BoxScaleLoss=1.539,ClassLoss=0.644
INFO:root:[Epoch 4] Training cost: 5.116, ObjLoss=7.835,BoxCenterLoss=3.812,BoxScaleLoss=1.476,ClassLoss=0.714
INFO:root:[Epoch 5] Training cost: 7.344, ObjLoss=6.980,BoxCenterLoss=3.646,BoxScaleLoss=1.266,ClassLoss=0.574
INFO:root:[Epoch 6] Training cost: 7.595, ObjLoss=6.505,BoxCenterLoss=3.471,BoxScaleLoss=1.134,ClassLoss=0.477
INFO:root:[Epoch 7] Training cost: 4.916, ObjLoss=6.412,BoxCenterLoss=3.677,BoxScaleLoss=1.178,ClassLoss=0.555
INFO:root:[Epoch 8] Training cost: 7.097, ObjLoss=6.043,BoxCenterLoss=3.763,BoxScaleLoss=1.089,ClassLoss=0.454
INFO:root:[Epoch 9] Training cost: 6.154, ObjLoss=5.690,BoxCenterLoss=3.519,BoxScaleLoss=1.110,ClassLoss=0.422
INFO:root:[Epoch 10] Training cost: 4.791, ObjLoss=5.472,BoxCenterLoss=3.662,BoxScaleLoss=1.072,ClassLoss=0.453
INFO:root:[Epoch 11] Training cost: 5.970, ObjLoss=5.234,BoxCenterLoss=3.605,BoxScaleLoss=1.059,ClassLoss=0.411
INFO:root:[Epoch 12] Training cost: 6.086, ObjLoss=4.999,BoxCenterLoss=3.718,BoxScaleLoss=1.040,ClassLoss=0.392
INFO:root:[Epoch 13] Training cost: 8.256, ObjLoss=5.373,BoxCenterLoss=3.504,BoxScaleLoss=0.941,ClassLoss=0.315
INFO:root:[Epoch 14] Training cost: 6.383, ObjLoss=5.360,BoxCenterLoss=3.724,BoxScaleLoss=1.014,ClassLoss=0.353
INFO:root:[Epoch 15] Training cost: 5.257, ObjLoss=5.411,BoxCenterLoss=3.624,BoxScaleLoss=1.036,ClassLoss=0.351
INFO:root:[Epoch 16] Training cost: 3.830, ObjLoss=4.875,BoxCenterLoss=3.457,BoxScaleLoss=0.917,ClassLoss=0.363
INFO:root:[Epoch 17] Training cost: 4.103, ObjLoss=4.808,BoxCenterLoss=3.487,BoxScaleLoss=0.923,ClassLoss=0.393
INFO:root:[Epoch 18] Training cost: 7.804, ObjLoss=5.314,BoxCenterLoss=3.476,BoxScaleLoss=0.958,ClassLoss=0.271
INFO:root:[Epoch 19] Training cost: 4.236, ObjLoss=5.220,BoxCenterLoss=3.606,BoxScaleLoss=1.041,ClassLoss=0.365
INFO:root:[Epoch 20] Training cost: 5.314, ObjLoss=4.575,BoxCenterLoss=3.587,BoxScaleLoss=0.971,ClassLoss=0.326
INFO:root:[Epoch 21] Training cost: 6.062, ObjLoss=4.537,BoxCenterLoss=3.535,BoxScaleLoss=0.888,ClassLoss=0.278
INFO:root:[Epoch 22] Training cost: 5.910, ObjLoss=4.299,BoxCenterLoss=3.287,BoxScaleLoss=0.824,ClassLoss=0.256
INFO:root:[Epoch 23] Training cost: 5.279, ObjLoss=4.495,BoxCenterLoss=3.482,BoxScaleLoss=0.886,ClassLoss=0.277
INFO:root:[Epoch 24] Training cost: 6.433, ObjLoss=4.542,BoxCenterLoss=3.560,BoxScaleLoss=0.816,ClassLoss=0.215
INFO:root:[Epoch 25] Training cost: 7.763, ObjLoss=4.627,BoxCenterLoss=3.465,BoxScaleLoss=0.874,ClassLoss=0.219
INFO:root:[Epoch 26] Training cost: 5.843, ObjLoss=4.568,BoxCenterLoss=3.552,BoxScaleLoss=0.890,ClassLoss=0.270
INFO:root:[Epoch 27] Training cost: 5.966, ObjLoss=4.022,BoxCenterLoss=3.189,BoxScaleLoss=0.818,ClassLoss=0.198
INFO:root:[Epoch 28] Training cost: 8.171, ObjLoss=4.460,BoxCenterLoss=3.438,BoxScaleLoss=0.867,ClassLoss=0.163
INFO:root:[Epoch 29] Training cost: 4.489, ObjLoss=4.575,BoxCenterLoss=3.582,BoxScaleLoss=0.854,ClassLoss=0.295

Note that num_trials=2 above is only used to speed up the tutorial. In normal practice, it is common to only use time_limits and drop num_trials. Also note that hyperparameter tuning defaults to random search. Model-based variants, such as search_strategy='bayesopt' or search_strategy='bayesopt_hyperband' can be a lot more sample-efficient.

After fitting, AutoGluon automatically returns the best model among all models in the searching space. From the output, we know the best model is the one trained with the second learning rate. To see how well the returned model performed on test dataset, call detector.evaluate().

dataset_test = task.Dataset(data_root, index_file_name='test', classes=('motorbike',))

test_map = detector.evaluate(dataset_test)
print("mAP on test dataset: {}".format(test_map[1][1]))

mAP on test dataset: 0.799905155044474

Below, we randomly select an image from test dataset and show the predicted box and probability over the origin image.

image = '000467.jpg'
image_path = os.path.join(data_root, 'JPEGImages', image)

ind, prob, loc = detector.predict(image_path)

../../_images/output_beginner_1fd7a6_11_0.png

We can also save the trained model, and use it later.

savefile = 'model.pkl'
detector.save(savefile)

from autogluon.vision import Detector
new_detector = Detector.load(savefile)