AutoMM Detection - Prepare Pascal VOC Dataset¶
Pascal VOC is a collection of datasets for object detection. The most commonly combination for benchmarking is using VOC2007 trainval and VOC2012 trainval for training and VOC2007 test for validation. Both VOC2007 and VOC2012 have the same 20 classes, and they have 16551 training images in total. This tutorial will walk through the steps of preparing both VOC2007 and VOC2012 for Autogluon MultiModalPredictor.
You need 8.4 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 10 min on AWS EC2 with EBS.
VOC has an official webpage to download the data,
but it’s always easier to perform a one-step setup.
We prepared a script to download both VOC2007 and VOC2012 in our examples:
download_voc0712.sh.
You can also download them separately:
download_voc07.sh,
download_voc12.sh.
Or you can also use our cli tool prepare_detection_dataset
that can download all datasets mentioned in our tutorials.
This python script is in our code:
prepare_detection_dataset.py,
and you can also run it as a cli: python3 -m autogluon.multimodal.cli.prepare_detection_dataset
.
Download with Python Script¶
The python script does not show progress bar, but is promised to work on all major platforms. If you are working on a Unix system and needs a progress bar, try the bash script!
You could either extract it under current directory by running:
python3 -m autogluon.multimodal.cli.prepare_detection_dataset --dataset_name voc0712
or extract it under a provided output path:
python3 -m autogluon.multimodal.cli.prepare_detection_dataset --dataset_name voc0712 --output_path ~/data
or make it shorter:
python3 -m autogluon.multimodal.cli.prepare_detection_dataset -d voc -o ~/data
or download them separately
python3 -m autogluon.multimodal.cli.prepare_detection_dataset -d voc07 -o ~/data
python3 -m autogluon.multimodal.cli.prepare_detection_dataset -d voc12 -o ~/data
Download with Bash Script¶
You could either extract it under current directory by running:
bash download_voc0712.sh
or extract it under a provided output path:
bash download_voc0712.sh ~/data
The command line output will show the progress bar:
extract data in current directory
Downloading VOC2007 trainval ...
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 438M 100 438M 0 0 92.3M 0 0:00:04 0:00:04 --:--:-- 95.5M
Downloading VOC2007 test data ...
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 430M 100 430M 0 0 96.5M 0 0:00:04 0:00:04 --:--:-- 99.1M
Downloading VOC2012 trainval ...
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
73 1907M 73 1401M 0 0 108M 0 0:00:17 0:00:12 0:00:05 118M
And after it finished, VOC datasets are extracted in folder VOCdevkit
, it contains
VOC2007 VOC2012
And both of them contains:
Annotations ImageSets JPEGImages SegmentationClass SegmentationObject
The VOC Format¶
VOC also refers to the specific format (in .xml
file) the VOC dataset is using.
In Autogluon MultiModalPredictor, we strongly recommend using COCO as your data format instead. Check AutoMM Detection - Prepare COCO2017 Dataset and Convert Data to COCO Format for more information about COCO dataset and how to convert a VOC dataset to COCO.
However, for fast proof testing we also have limit support for VOC format. While using VOC format dataset, the input is the root path of the dataset, and contains at least:
Annotations ImageSets JPEGImages
Other Examples¶
You may go to AutoMM Examples to explore other examples about AutoMM.
Customization¶
To learn how to customize AutoMM, please refer to Customize AutoMM.
Citation¶
@Article{Everingham10,
author = "Everingham, M. and Van~Gool, L. and Williams, C. K. I. and Winn, J. and Zisserman, A.",
title = "The Pascal Visual Object Classes (VOC) Challenge",
journal = "International Journal of Computer Vision",
volume = "88",
year = "2010",
number = "2",
month = jun,
pages = "303--338",
}