CVEDIA is a free, cloud-based service that simplifies image dataset preparation and management.
You can use it to access standardized versions of public image datasets such as Open Images, COCO, ImageNet, and SpaceNet. Using CVEDIA's open-source CLI tool, you can easily export your filters and augmentations to your local server in the format of your favorite machine learning framework.
The process of creating a dataset, annotationg it, and exporting it to a machine learning framework features the following steps:
A dataset contains all visual data, meta data and subsets. You can create a dataset from a existing one, a derivative, or upload your own. It's also possible to create a hybrid from a derivative as you can upload your own content after creating it. For more information, see Datasets.
After dataset creation you can start adding and refining annotations. Annotations are bound to a layer and label system, allowing them to be organized and categorized. Labels always have a color bound to it, and can contain abritrary values that can be later exported. For more information, see Adding annotations.
Augmentation allows one to add arbitrary transformations on images. This a great way to incrase entropy on your images without adding any new annotations. CVEDIA supports over 20 different types of augmentations that can be stacked and configured in many ways, including random parameters that can create a stream of endless variations of a image. For more information, see Augmenting images.
Consider that the augmentation process can output images on your kernel size, avoiding resizes on CPU level before feeding the model and requiring almost no storage to begin training.
Using our CVEDIA's open source CLI tool you will be able to export the augmentation you created to your local disk, with optional formats supported by Cafee and Tensorflow, requiring no data tranformation. For more information, see Exporting datasets.