Table Of Contents
Table Of Contents

Prepare COCO datasets

COCO is a large-scale object detection, segmentation, and captioning datasetself. This tutorial will walk through the steps of preparing this dataset for GluonCV.

http://cocodataset.org/images/coco-logo.png

Hint

You need 42.7 GB disk space to download and extract this dataset. SSD is preferred over HDD because of its better performance.

The total time to prepare the dataset depends on your Internet speed and disk performance. For example, it often takes 20 min on AWS EC2 with EBS.

Prepare the dataset

We need the following four files from COCO:

Filename Size SHA-1
train2017.zip 18 GB 10ad623668ab00c62c096f0ed636d6aff41faca5
val2017.zip 778 MB 4950dc9d00dbe1c933ee0170f5797584351d2a41
annotations_trainval2017.zip 241 MB 8551ee4bb5860311e79dace7e79cb91e432e78b3
stuff_annotations_trainval2017.zip 401 MB e7aa0f7515c07e23873a9f71d9095b06bcea3e12

The easiest way to download and unpack these files is to download helper script mscoco.py and run the following command:

python mscoco.py

which will automatically download and extract the data into ~/.mxnet/datasets/coco.

If you already have the above files sitting on your disk, you can set --download-dir to point to them. For example, assuming the files are saved in ~/coco/, you can run:

python mscoco.py --download-dir ~/coco

Read with GluonCV

Loading images and labels is straight-forward with gluoncv.data.COCODetection.

from gluoncv import data, utils
from matplotlib import pyplot as plt

train_dataset = data.COCODetection(splits=['instances_train2017'])
val_dataset = data.COCODetection(splits=['instances_val2017'])
print('Num of training images:', len(train_dataset))
print('Num of validation images:', len(val_dataset))

Out:

loading annotations into memory...
Done (t=14.78s)
creating index...
index created!
loading annotations into memory...
Done (t=0.43s)
creating index...
index created!
Num of training images: 117266
Num of validation images: 4952

Now let’s visualize one example.

train_image, train_label = train_dataset[0]
bounding_boxes = train_label[:, :4]
class_ids = train_label[:, 4:5]
print('Image size (height, width, RGB):', train_image.shape)
print('Num of objects:', bounding_boxes.shape[0])
print('Bounding boxes (num_boxes, x_min, y_min, x_max, y_max):\n',
      bounding_boxes)
print('Class IDs (num_boxes, ):\n', class_ids)

utils.viz.plot_bbox(train_image.asnumpy(), bounding_boxes, scores=None,
                    labels=class_ids, class_names=train_dataset.classes)
plt.show()
../../_images/sphx_glr_mscoco_001.png

Out:

Image size (height, width, RGB): (480, 640, 3)
Num of objects: 8
Bounding boxes (num_boxes, x_min, y_min, x_max, y_max):
 [[  1.08 187.69 611.67 472.53]
 [311.73   4.31 630.01 231.99]
 [249.6  229.27 564.84 473.35]
 [  0.    13.51 433.48 387.63]
 [376.2   40.36 450.75  85.89]
 [465.78  38.97 522.85  84.64]
 [385.7   73.66 468.72 143.17]
 [364.05   2.49 457.81  72.56]]
Class IDs (num_boxes, ):
 [[45.]
 [45.]
 [50.]
 [45.]
 [49.]
 [49.]
 [49.]
 [49.]]

Finally, to use both train_dataset and val_dataset for training, we can pass them through data transformations and load with mxnet.gluon.data.DataLoader, see train_ssd.py for more information.

Total running time of the script: ( 0 minutes 40.075 seconds)

Gallery generated by Sphinx-Gallery