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 Prediction - 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 ObjectDetector:

from autogluon.vision import ObjectDetector
/var/lib/jenkins/workspace/workspace/autogluon-tutorial-object-detection-v3/venv/lib/python3.7/site-packages/gluoncv/__init__.py:40: UserWarning: Both mxnet==1.7.0 and torch==1.9.0+cu102 are installed. You might encounter increased GPU memory footprint if both framework are used at the same time.
  warnings.warn(f'Both mxnet=={mx.__version__} and torch=={torch.__version__} are installed. '

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 name of unzipped folder is called tiny_motorbike. Anyway, the task dataset helper can perform the download and extraction automatically, and load the dataset according to the detection formats.

url = 'https://autogluon.s3.amazonaws.com/datasets/tiny_motorbike.zip'
dataset_train = ObjectDetector.Dataset.from_voc(url, splits='trainval')
tiny_motorbike/
├── Annotations/
├── ImageSets/
└── JPEGImages/

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 5 epochs to avoid bursting our tutorial runtime.

time_limit = 60*30  # at most 0.5 hour
detector = ObjectDetector()
hyperparameters = {'epochs': 5, 'batch_size': 8}
hyperparameter_tune_kwargs={'num_trials': 2}
detector.fit(dataset_train, time_limit=time_limit, hyperparameters=hyperparameters, hyperparameter_tune_kwargs=hyperparameter_tune_kwargs)
The number of requested GPUs is greater than the number of available GPUs.Reduce the number to 1
Randomly split train_data into train[150]/validation[20] splits.
Starting HPO experiments
  0%|          | 0/2 [00:00<?, ?it/s]
modified configs(<old> != <new>): {
root.num_workers     4 != 8
root.dataset_root    ~/.mxnet/datasets/ != auto
root.dataset         voc_tiny != auto
root.train.batch_size 16 != 8
root.train.early_stop_baseline 0.0 != -inf
root.train.early_stop_max_value 1.0 != inf
root.train.early_stop_patience -1 != 10
root.train.seed      233 != 715
root.train.epochs    20 != 5
root.gpus            (0, 1, 2, 3) != (0,)
root.valid.batch_size 16 != 8
root.ssd.base_network vgg16_atrous != resnet50_v1
root.ssd.data_shape  300 != 512
}
Saved config to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-object-detection-v3/docs/_build/eval/tutorials/object_detection/8b829587/.trial_0/config.yaml
Using transfer learning from ssd_512_resnet50_v1_coco, the other network parameters are ignored.
Start training from [Epoch 0]
[Epoch 0] Training cost: 8.862641, CrossEntropy=3.547565, SmoothL1=0.999755
[Epoch 0] Validation:
person=0.6550085253092772
cow=nan
bus=1.0000000000000002
bicycle=0.5000000000000001
dog=0.0
chair=nan
boat=nan
motorbike=0.7510034809067803
car=1.0000000000000002
pottedplant=0.0
mAP=0.5580017151737225
[Epoch 0] Current best map: 0.558002 vs previous 0.000000, saved to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-object-detection-v3/docs/_build/eval/tutorials/object_detection/8b829587/.trial_0/best_checkpoint.pkl
[Epoch 1] Training cost: 8.162877, CrossEntropy=2.745820, SmoothL1=1.159730
[Epoch 1] Validation:
person=0.7453687408338571
cow=nan
bus=1.0000000000000002
bicycle=0.03636363636363636
dog=1.0000000000000002
chair=nan
boat=nan
motorbike=0.7745454545454544
car=1.0000000000000002
pottedplant=0.0
mAP=0.6508968331061354
[Epoch 1] Current best map: 0.650897 vs previous 0.558002, saved to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-object-detection-v3/docs/_build/eval/tutorials/object_detection/8b829587/.trial_0/best_checkpoint.pkl
[Epoch 2] Training cost: 8.205384, CrossEntropy=2.389947, SmoothL1=1.118457
[Epoch 2] Validation:
person=0.8148368996407662
cow=nan
bus=1.0000000000000002
bicycle=0.03896103896103896
dog=0.33333333333333326
chair=nan
boat=nan
motorbike=0.833822091886608
car=1.0000000000000002
pottedplant=0.0
mAP=0.5744219091173923
[Epoch 3] Training cost: 8.239273, CrossEntropy=2.275549, SmoothL1=0.931290
[Epoch 3] Validation:
person=0.8376956617222866
cow=nan
bus=1.0000000000000002
bicycle=0.0
dog=1.0000000000000002
chair=nan
boat=nan
motorbike=0.7817730838067444
car=1.0000000000000002
pottedplant=0.0
mAP=0.6599241065041473
[Epoch 3] Current best map: 0.659924 vs previous 0.650897, saved to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-object-detection-v3/docs/_build/eval/tutorials/object_detection/8b829587/.trial_0/best_checkpoint.pkl
[Epoch 4] Training cost: 7.828332, CrossEntropy=2.373933, SmoothL1=1.050903
[Epoch 4] Validation:
person=0.7006795973767732
cow=nan
bus=1.0000000000000002
bicycle=0.028708133971291867
dog=0.0
chair=nan
boat=nan
motorbike=0.7874027825102737
car=0.25000000000000006
pottedplant=0.0
mAP=0.3952557876940484
Applying the state from the best checkpoint...
modified configs(<old> != <new>): {
root.num_workers     4 != 8
root.dataset_root    ~/.mxnet/datasets/ != auto
root.dataset         voc_tiny != auto
root.train.epochs    20 != 5
root.train.batch_size 16 != 8
root.train.early_stop_baseline 0.0 != -inf
root.train.early_stop_max_value 1.0 != inf
root.train.early_stop_patience -1 != 10
root.train.seed      233 != 715
root.gpus            (0, 1, 2, 3) != (0,)
root.valid.batch_size 16 != 8
}
Saved config to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-object-detection-v3/docs/_build/eval/tutorials/object_detection/8b829587/.trial_1/config.yaml
Using transfer learning from yolo3_darknet53_coco, the other network parameters are ignored.
Start training from [Epoch 0]
[Epoch 0] Training cost: 13.318, ObjLoss=8.721, BoxCenterLoss=7.626, BoxScaleLoss=2.039, ClassLoss=4.037
[Epoch 0] Validation:
person=0.544292643673739
cow=nan
bus=1.0000000000000002
bicycle=1.0000000000000002
dog=0.0
chair=nan
boat=nan
motorbike=0.7931228500146169
car=0.5000000000000001
pottedplant=0.0
mAP=0.548202213384051
[Epoch 0] Current best map: 0.548202 vs previous 0.000000, saved to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-object-detection-v3/docs/_build/eval/tutorials/object_detection/8b829587/.trial_1/best_checkpoint.pkl
[Epoch 1] Training cost: 8.208, ObjLoss=8.763, BoxCenterLoss=7.635, BoxScaleLoss=2.637, ClassLoss=3.511
[Epoch 1] Validation:
person=0.5122707801569751
cow=nan
bus=1.0000000000000002
bicycle=0.6666666666666665
dog=0.5000000000000001
chair=nan
boat=nan
motorbike=0.817133520560294
car=0.5000000000000001
pottedplant=0.0
mAP=0.570867281054848
[Epoch 1] Current best map: 0.570867 vs previous 0.548202, saved to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-object-detection-v3/docs/_build/eval/tutorials/object_detection/8b829587/.trial_1/best_checkpoint.pkl
[Epoch 2] Training cost: 11.691, ObjLoss=9.229, BoxCenterLoss=7.855, BoxScaleLoss=2.943, ClassLoss=3.415
[Epoch 2] Validation:
person=0.4699096225412015
cow=nan
bus=1.0000000000000002
bicycle=0.4000000000000001
dog=0.0
chair=nan
boat=nan
motorbike=0.6385327664559888
car=1.0000000000000002
pottedplant=0.0
mAP=0.5012060555710273
[Epoch 3] Training cost: 14.954, ObjLoss=9.605, BoxCenterLoss=7.870, BoxScaleLoss=3.045, ClassLoss=3.250
[Epoch 3] Validation:
person=0.671712976945535
cow=nan
bus=1.0000000000000002
bicycle=1.0000000000000002
dog=0.0
chair=nan
boat=nan
motorbike=0.7820218996689585
car=0.5000000000000001
pottedplant=0.0
mAP=0.5648192680877848
[Epoch 4] Training cost: 12.014, ObjLoss=9.704, BoxCenterLoss=7.974, BoxScaleLoss=3.082, ClassLoss=3.045
[Epoch 4] Validation:
person=0.8149235857310393
cow=nan
bus=1.0000000000000002
bicycle=0.05454545454545456
dog=0.0
chair=nan
boat=nan
motorbike=0.6814861275088547
car=1.0000000000000002
pottedplant=0.0
mAP=0.5072793096836213
Applying the state from the best checkpoint...
modified configs(<old> != <new>): {
root.num_workers     4 != 8
root.dataset_root    ~/.mxnet/datasets/ != auto
root.dataset         voc_tiny != auto
root.train.epochs    20 != 5
root.train.batch_size 16 != 8
root.train.early_stop_baseline 0.0 != -inf
root.train.early_stop_max_value 1.0 != inf
root.train.early_stop_patience -1 != 10
root.train.seed      233 != 715
root.gpus            (0, 1, 2, 3) != (0,)
root.valid.batch_size 16 != 8
}
Saved config to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-object-detection-v3/docs/_build/eval/tutorials/object_detection/8b829587/.trial_0/config.yaml
Using transfer learning from yolo3_darknet53_coco, the other network parameters are ignored.
Start training from [Epoch 0]
[Epoch 0] Training cost: 11.484, ObjLoss=10.139, BoxCenterLoss=7.506, BoxScaleLoss=2.054, ClassLoss=4.329
[Epoch 0] Validation:
person=0.5953604135422318
cow=nan
bus=1.0000000000000002
bicycle=1.0000000000000002
dog=1.0000000000000002
chair=nan
boat=nan
motorbike=0.6650385833096322
car=0.25000000000000006
pottedplant=0.0
mAP=0.6443427138359806
[Epoch 0] Current best map: 0.644343 vs previous 0.000000, saved to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-object-detection-v3/docs/_build/eval/tutorials/object_detection/8b829587/.trial_0/best_checkpoint.pkl
[Epoch 1] Training cost: 8.047, ObjLoss=9.730, BoxCenterLoss=7.472, BoxScaleLoss=2.511, ClassLoss=3.714
[Epoch 1] Validation:
person=0.4387786531230103
cow=nan
bus=1.0000000000000002
bicycle=0.5454545454545455
dog=0.0
chair=nan
boat=nan
motorbike=0.7694829244829245
car=0.09090909090909091
pottedplant=0.0
mAP=0.40637503056708163
[Epoch 2] Training cost: 11.591, ObjLoss=10.031, BoxCenterLoss=7.653, BoxScaleLoss=3.022, ClassLoss=3.371
[Epoch 2] Validation:
person=0.7086336371095728
cow=nan
bus=1.0000000000000002
bicycle=0.5454545454545455
dog=1.0000000000000002
chair=nan
boat=nan
motorbike=0.7479565784745732
car=0.0
pottedplant=0.0
mAP=0.5717206801483846
[Epoch 3] Training cost: 15.022, ObjLoss=10.499, BoxCenterLoss=7.891, BoxScaleLoss=3.236, ClassLoss=3.188
[Epoch 3] Validation:
person=0.6266646963498219
cow=nan
bus=1.0000000000000002
bicycle=1.0000000000000002
dog=0.0
chair=nan
boat=nan
motorbike=0.8665926262630411
car=0.33333333333333326
pottedplant=0.0
mAP=0.5466558079923137
[Epoch 4] Training cost: 11.700, ObjLoss=10.314, BoxCenterLoss=7.912, BoxScaleLoss=3.251, ClassLoss=2.969
[Epoch 4] Validation:
person=0.5068840579710145
cow=nan
bus=1.0000000000000002
bicycle=1.0000000000000002
dog=0.0
chair=nan
boat=nan
motorbike=0.8322394451426711
car=0.5000000000000001
pottedplant=0.0
mAP=0.5484462147305266
Applying the state from the best checkpoint...
Finished, total runtime is 234.83 s
{ 'best_config': { 'dataset': 'auto',
                   'dataset_root': 'auto',
                   'estimator': <class 'gluoncv.auto.estimators.yolo.yolo.YOLOv3Estimator'>,
                   'gpus': [0],
                   'horovod': False,
                   'num_workers': 8,
                   'resume': '',
                   'save_interval': 10,
                   'save_prefix': '',
                   'train': { 'batch_size': 8,
                              'early_stop_baseline': -inf,
                              'early_stop_max_value': inf,
                              'early_stop_min_delta': 0.001,
                              'early_stop_patience': 10,
                              'epochs': 5,
                              'label_smooth': False,
                              'log_interval': 100,
                              'lr': 0.001,
                              'lr_decay': 0.1,
                              'lr_decay_epoch': (160, 180),
                              'lr_decay_period': 0,
                              'lr_mode': 'step',
                              'mixup': False,
                              'momentum': 0.9,
                              'no_mixup_epochs': 20,
                              'no_wd': False,
                              'num_samples': -1,
                              'seed': 715,
                              'start_epoch': 0,
                              'warmup_epochs': 0,
                              'warmup_lr': 0.0,
                              'wd': 0.0005},
                   'valid': { 'batch_size': 8,
                              'iou_thresh': 0.5,
                              'metric': 'voc07',
                              'val_interval': 1},
                   'yolo3': { 'amp': False,
                              'anchors': ( [10, 13, 16, 30, 33, 23],
                                           [30, 61, 62, 45, 59, 119],
                                           [116, 90, 156, 198, 373, 326]),
                              'base_network': 'darknet53',
                              'data_shape': 416,
                              'filters': (512, 256, 128),
                              'nms_thresh': 0.45,
                              'nms_topk': 400,
                              'no_random_shape': False,
                              'strides': (8, 16, 32),
                              'syncbn': False,
                              'transfer': 'yolo3_darknet53_coco'}},
  'total_time': 234.83203172683716,
  'train_map': 0.5028437862370443,
  'valid_map': 0.6443427138359806}
<autogluon.vision.detector.detector.ObjectDetector at 0x7fc5587c3310>

Note that num_trials=2 above is only used to speed up the tutorial. In normal practice, it is common to only use time_limit and drop num_trials. Also note that hyperparameter tuning defaults to random search. Model-based variants, such as searcher='bayesopt' in hyperparameter_tune_kwargs 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 = ObjectDetector.Dataset.from_voc(url, splits='test')

test_map = detector.evaluate(dataset_test)
print("mAP on test dataset: {}".format(test_map[1][-1]))
tiny_motorbike/
├── Annotations/
├── ImageSets/
└── JPEGImages/
mAP on test dataset: 0.1456443656975456

Below, we randomly select an image from test dataset and show the predicted class, box and probability over the origin image, stored in predict_class, predict_rois and predict_score columns, respectively. You can interpret predict_rois as a dict of (xmin, ymin, xmax, ymax) proportional to original image size.

image_path = dataset_test.iloc[0]['image']
result = detector.predict(image_path)
print(result)
   predict_class  predict_score  0         person       0.670992
1      motorbike       0.548376
2      motorbike       0.465958
3         person       0.297391
4         person       0.263531
5         person       0.168833
6         person       0.101813
7      motorbike       0.080886
8      motorbike       0.073280
9            car       0.073177
10     motorbike       0.048323
11     motorbike       0.047944
12     motorbike       0.033986
13   pottedplant       0.022304
14     motorbike       0.019886
15   pottedplant       0.018582
16           car       0.018488
17     motorbike       0.017762
18        person       0.013835
19   pottedplant       0.013085
20           dog       0.012839
21     motorbike       0.012794
22        person       0.012009
23     motorbike       0.011984
24       bicycle       0.011897
25        person       0.010887
26        person       0.010562
27        person       0.010405

                                         predict_rois
0   {'xmin': 0.3782995939254761, 'ymin': 0.2945322...
1   {'xmin': 0.0, 'ymin': 0.6371848583221436, 'xma...
2   {'xmin': 0.31195250153541565, 'ymin': 0.452559...
3   {'xmin': 0.6325993537902832, 'ymin': 0.0435023...
4   {'xmin': 0.7543973326683044, 'ymin': 0.0450115...
5   {'xmin': 0.8843181729316711, 'ymin': 0.0114150...
6   {'xmin': 0.5207630395889282, 'ymin': 0.0280970...
7   {'xmin': 0.0361197255551815, 'ymin': 0.4919169...
8   {'xmin': 0.38730472326278687, 'ymin': 0.318395...
9   {'xmin': 0.0361197255551815, 'ymin': 0.4919169...
10  {'xmin': 0.7543973326683044, 'ymin': 0.0450115...
11  {'xmin': 0.6325993537902832, 'ymin': 0.0435023...
12  {'xmin': 0.0, 'ymin': 0.0, 'xmax': 1.0, 'ymax'...
13  {'xmin': 0.0, 'ymin': 0.6371848583221436, 'xma...
14  {'xmin': 0.5207630395889282, 'ymin': 0.0280970...
15  {'xmin': 0.3782995939254761, 'ymin': 0.2945322...
16  {'xmin': 0.0, 'ymin': 0.6519644856452942, 'xma...
17  {'xmin': 0.8843181729316711, 'ymin': 0.0114150...
18  {'xmin': 0.6730436086654663, 'ymin': 0.0291886...
19  {'xmin': 0.6325993537902832, 'ymin': 0.0435023...
20  {'xmin': 0.31195250153541565, 'ymin': 0.452559...
21  {'xmin': 0.011701270937919617, 'ymin': 0.03621...
22  {'xmin': 0.5618847012519836, 'ymin': 0.0097288...
23  {'xmin': 0.7213250398635864, 'ymin': 0.3840133...
24  {'xmin': 0.31195250153541565, 'ymin': 0.452559...
25  {'xmin': 0.3077634274959564, 'ymin': 0.3562027...
26  {'xmin': 0.41763123869895935, 'ymin': 0.272201...
27  {'xmin': 0.45487532019615173, 'ymin': 0.013570...

Prediction with multiple images is permitted:

bulk_result = detector.predict(dataset_test)
print(bulk_result)
     predict_class  predict_score  0           person       0.670992
1        motorbike       0.548376
2        motorbike       0.465958
3           person       0.297391
4           person       0.263531
...            ...            ...
1828     motorbike       0.033558
1829        person       0.031899
1830        person       0.017892
1831        person       0.012938
1832     motorbike       0.012641

                                           predict_rois  0     {'xmin': 0.3782995939254761, 'ymin': 0.2945322...
1     {'xmin': 0.0, 'ymin': 0.6371848583221436, 'xma...
2     {'xmin': 0.31195250153541565, 'ymin': 0.452559...
3     {'xmin': 0.6325993537902832, 'ymin': 0.0435023...
4     {'xmin': 0.7543973326683044, 'ymin': 0.0450115...
...                                                 ...
1828  {'xmin': 0.27905958890914917, 'ymin': 0.137294...
1829  {'xmin': 0.29887059330940247, 'ymin': 0.221315...
1830  {'xmin': 0.295794278383255, 'ymin': 0.05811200...
1831  {'xmin': 0.018482832238078117, 'ymin': 0.46835...
1832  {'xmin': 0.018482832238078117, 'ymin': 0.46835...

                                                  image
0     /var/lib/jenkins/.gluoncv/datasets/tiny_motorb...
1     /var/lib/jenkins/.gluoncv/datasets/tiny_motorb...
2     /var/lib/jenkins/.gluoncv/datasets/tiny_motorb...
3     /var/lib/jenkins/.gluoncv/datasets/tiny_motorb...
4     /var/lib/jenkins/.gluoncv/datasets/tiny_motorb...
...                                                 ...
1828  /var/lib/jenkins/.gluoncv/datasets/tiny_motorb...
1829  /var/lib/jenkins/.gluoncv/datasets/tiny_motorb...
1830  /var/lib/jenkins/.gluoncv/datasets/tiny_motorb...
1831  /var/lib/jenkins/.gluoncv/datasets/tiny_motorb...
1832  /var/lib/jenkins/.gluoncv/datasets/tiny_motorb...

[1833 rows x 4 columns]

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

savefile = 'detector.ag'
detector.save(savefile)
new_detector = ObjectDetector.load(savefile)