Prepare PASCAL VOC datasets¶
Pascal VOC contains a collection of datasets for object detection. The most commonly adopted version for benchmarking is using 2007 trainval and 2012 trainval for training and 2007 test for validation. This tutorial will walk you through the steps for preparing this dataset to be used by GluonVision.
Prepare the dataset¶
The easiest way is simply running this script:
which will automatically download and extract the data into ~/.mxnet/datasets/voc.
python pascal_voc.py
Note
You need 8.4 GB disk space to download and extract this dataset. SSD is preferred over HDD because of its better performance.
Note
The total time to prepare the dataset depends on your Internet speed and disk performance. For example, it often takes 10min on AWS EC2 with EBS.
If you have already downloaded the following required files
| Filename | Size | SHA-1 |
|---|---|---|
| VOCtrainval_06-Nov-2007.tar | 439 MB | 34ed68851bce2a36e2a223fa52c661d592c66b3c |
| VOCtest_06-Nov-2007.tar | 430 MB | 41a8d6e12baa5ab18ee7f8f8029b9e11805b4ef1 |
| VOCtrainval_11-May-2012.tar | 1.9 GB | 4e443f8a2eca6b1dac8a6c57641b67dd40621a49 |
| benchmark.tgz | 1.4 GB | 7129e0a480c2d6afb02b517bb18ac54283bfaa35 |
then you can specify the folder name through --dir to avoid
download them again.
For example, make sure you have these files exist in ~/VOCdevkit/downloads, and you can run
python pascal_voc.py --dir ~/VOCdevkit
to extract them.
How to load the dataset¶
Load image and label from Pascal VOC is quite straight-forward
from gluonvision.data import VOCDetection
train_dataset = VOCDetection(splits=[(2007, 'trainval'), (2012, 'trainval')])
val_dataset = VOCDetection(splits=[(2007, 'test')])
print('Training images:', len(train_dataset))
print('Validation images:', len(val_dataset))
Out:
Training images: 16551
Validation images: 4952
Check the first example¶
train_image, train_label = train_dataset[0]
bboxes = train_label[:, :4]
cids = train_label[:, 4:5]
print('image size:', train_image.shape)
print('bboxes:', bboxes.shape, 'class ids:', cids.shape)
from matplotlib import pyplot as plt
from gluonvision.utils import viz
ax = viz.plot_bbox(train_image.asnumpy(), bboxes, scores=None, labels=cids, class_names=train_dataset.classes)
plt.show()
Out:
image size: (375, 500, 3)
bboxes: (5, 4) class ids: (5, 1)