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/miniconda3/envs/autogluon-tutorial-object-detection-v3/lib/python3.9/site-packages/gluoncv/__init__.py:40: UserWarning: Both mxnet==1.9.0 and torch==1.10.2+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)
=============================================================================
WARNING: ObjectDetector is deprecated as of v0.4.0 and may contain various bugs and issues!
In a future release ObjectDetector may be entirely reworked to use Torch as a backend.
This future change will likely be API breaking.Users should ensure they update their code that depends on ObjectDetector when upgrading to future AutoGluon releases.
For more information, refer to ObjectDetector refactor GitHub issue: https://github.com/awslabs/autogluon/issues/1559
=============================================================================

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.gpus            (0, 1, 2, 3) != (0,)
root.dataset         voc_tiny != auto
root.train.epochs    20 != 5
root.train.early_stop_max_value 1.0 != inf
root.train.early_stop_patience -1 != 10
root.train.early_stop_baseline 0.0 != -inf
root.train.batch_size 16 != 8
root.train.seed      233 != 737
root.ssd.base_network vgg16_atrous != resnet50_v1
root.ssd.data_shape  300 != 512
root.dataset_root    ~/.mxnet/datasets/ != auto
root.num_workers     4 != 8
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/4da478bb/.trial_0/config.yaml
No gpu detected, fallback to cpu. You can ignore this warning if this is intended.
Using transfer learning from ssd_512_resnet50_v1_coco, the other network parameters are ignored.
Start training from [Epoch 0]
[Epoch 0] Training cost: 65.590887, CrossEntropy=3.801192, SmoothL1=1.055731
[Epoch 0] Validation:
dog=nan
car=0.06818181818181819
bicycle=0.33333333333333326
person=0.7678063637767204
boat=nan
bus=nan
motorbike=0.6672188467270435
chair=nan
pottedplant=nan
cow=1.0000000000000002
mAP=0.5673080724037831
[Epoch 0] Current best map: 0.567308 vs previous 0.000000, saved to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-object-detection-v3/docs/_build/eval/tutorials/object_detection/4da478bb/.trial_0/best_checkpoint.pkl
[Epoch 1] Training cost: 68.930588, CrossEntropy=2.772563, SmoothL1=1.296152
[Epoch 1] Validation:
dog=nan
car=0.028708133971291867
bicycle=1.0000000000000002
person=0.8146370104990794
boat=nan
bus=nan
motorbike=0.8019951063429325
chair=nan
pottedplant=nan
cow=0.07792207792207792
mAP=0.5446524657470764
[Epoch 2] Training cost: 68.206152, CrossEntropy=2.479209, SmoothL1=1.082939
[Epoch 2] Validation:
dog=nan
car=1.0000000000000002
bicycle=1.0000000000000002
person=0.7458585858585858
boat=nan
bus=nan
motorbike=0.8579902180307861
chair=nan
pottedplant=nan
cow=0.03896103896103896
mAP=0.7285619685700822
[Epoch 2] Current best map: 0.728562 vs previous 0.567308, saved to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-object-detection-v3/docs/_build/eval/tutorials/object_detection/4da478bb/.trial_0/best_checkpoint.pkl
[Epoch 3] Training cost: 68.468571, CrossEntropy=2.209581, SmoothL1=1.019761
[Epoch 3] Validation:
dog=nan
car=0.7727272727272726
bicycle=0.5000000000000001
person=0.8263447079236554
boat=nan
bus=nan
motorbike=0.876943346508564
chair=nan
pottedplant=nan
cow=0.6363636363636365
mAP=0.7224757927046257
[Epoch 4] Training cost: 64.457107, CrossEntropy=2.044162, SmoothL1=0.928696
[Epoch 4] Validation:
dog=nan
car=0.7727272727272726
bicycle=1.0000000000000002
person=0.8244680752478918
boat=nan
bus=nan
motorbike=0.9029040404040406
chair=nan
pottedplant=nan
cow=0.7272727272727274
mAP=0.8454744231303865
[Epoch 4] Current best map: 0.845474 vs previous 0.728562, saved to /var/lib/jenkins/workspace/workspace/autogluon-tutorial-object-detection-v3/docs/_build/eval/tutorials/object_detection/4da478bb/.trial_0/best_checkpoint.pkl
Applying the state from the best checkpoint...
Finished, total runtime is 384.89 s
{ 'best_config': { 'dataset': 'auto',
                   'dataset_root': 'auto',
                   'estimator': <class 'gluoncv.auto.estimators.ssd.ssd.SSDEstimator'>,
                   'gpus': [0],
                   'horovod': False,
                   'num_workers': 8,
                   'resume': '',
                   'save_interval': 1,
                   'ssd': { 'amp': False,
                            'base_network': 'resnet50_v1',
                            'data_shape': 512,
                            'filters': None,
                            'nms_thresh': 0.45,
                            'nms_topk': 400,
                            'ratios': ( [1, 2, 0.5],
                                        [1, 2, 0.5, 3, 0.3333333333333333],
                                        [1, 2, 0.5, 3, 0.3333333333333333],
                                        [1, 2, 0.5, 3, 0.3333333333333333],
                                        [1, 2, 0.5],
                                        [1, 2, 0.5]),
                            'sizes': (30, 60, 111, 162, 213, 264, 315),
                            'steps': (8, 16, 32, 64, 100, 300),
                            'syncbn': False,
                            'transfer': 'ssd_512_resnet50_v1_coco'},
                   'train': { 'batch_size': 8,
                              'dali': False,
                              'early_stop_baseline': -inf,
                              'early_stop_max_value': inf,
                              'early_stop_min_delta': 0.001,
                              'early_stop_patience': 10,
                              'epochs': 5,
                              'log_interval': 100,
                              'lr': 0.001,
                              'lr_decay': 0.1,
                              'lr_decay_epoch': (160, 200),
                              'momentum': 0.9,
                              'seed': 737,
                              'start_epoch': 0,
                              'wd': 0.0005},
                   'valid': { 'batch_size': 8,
                              'iou_thresh': 0.5,
                              'metric': 'voc07',
                              'val_interval': 1}},
  'total_time': 384.8896007537842,
  'train_map': 0.7638803313469583,
  'valid_map': 0.8454744231303865}
<autogluon.vision.detector.detector.ObjectDetector at 0x7fdda74f56d0>

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.

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: 0.04450817589004445

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       0.991311
1      motorbike       0.982311
2            car       0.610832
3         person       0.130214
4         person       0.126884
..           ...            ...
95        person       0.021482
96        person       0.021342
97        person       0.021336
98        person       0.021281
99        person       0.021270

                                         predict_rois
0   {'xmin': 0.3961397409439087, 'ymin': 0.3010425...
1   {'xmin': 0.3202349841594696, 'ymin': 0.4307358...
2   {'xmin': 0.008988169953227043, 'ymin': 0.63452...
3   {'xmin': 0.9957932829856873, 'ymin': 0.5852979...
4   {'xmin': 0.9959208965301514, 'ymin': 0.7934117...
..                                                ...
95  {'xmin': 0.9940804839134216, 'ymin': 0.7975868...
96  {'xmin': 0.9985010027885437, 'ymin': 0.7635197...
97  {'xmin': 1.0, 'ymin': 0.5887960195541382, 'xma...
98  {'xmin': 0.9956215620040894, 'ymin': 0.3971858...
99  {'xmin': 0.9951635003089905, 'ymin': 0.5103151...

[100 rows x 3 columns]

Prediction with multiple images is permitted:

bulk_result = detector.predict(dataset_test)
print(bulk_result)
     predict_class  predict_score  0           person       0.991311
1        motorbike       0.982311
2              car       0.610832
3           person       0.130214
4           person       0.126884
...            ...            ...
4062         chair       0.017761
4063        person       0.017701
4064        person       0.017539
4065        person       0.017535
4066        person       0.017242

                                           predict_rois  0     {'xmin': 0.3961397409439087, 'ymin': 0.3010425...
1     {'xmin': 0.3202349841594696, 'ymin': 0.4307358...
2     {'xmin': 0.008988169953227043, 'ymin': 0.63452...
3     {'xmin': 0.9957932829856873, 'ymin': 0.5852979...
4     {'xmin': 0.9959208965301514, 'ymin': 0.7934117...
...                                                 ...
4062  {'xmin': 0.6061325073242188, 'ymin': 0.2420845...
4063  {'xmin': 0.9135922193527222, 'ymin': 0.7619636...
4064  {'xmin': 0.46921318769454956, 'ymin': 0.187138...
4065  {'xmin': 0.9625320434570312, 'ymin': 0.5775896...
4066  {'xmin': 0.9447792768478394, 'ymin': 0.7458862...

                                                  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...
...                                                 ...
4062  /var/lib/jenkins/.gluoncv/datasets/tiny_motorb...
4063  /var/lib/jenkins/.gluoncv/datasets/tiny_motorb...
4064  /var/lib/jenkins/.gluoncv/datasets/tiny_motorb...
4065  /var/lib/jenkins/.gluoncv/datasets/tiny_motorb...
4066  /var/lib/jenkins/.gluoncv/datasets/tiny_motorb...

[4067 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)
/var/lib/jenkins/miniconda3/envs/autogluon-tutorial-object-detection-v3/lib/python3.9/site-packages/mxnet/gluon/block.py:1784: UserWarning: Cannot decide type for the following arguments. Consider providing them as input:
    data: None
  input_sym_arg_type = in_param.infer_type()[0]