Nilvana Vision Studio - Dataset Versioning

Please follow the instructions to create different dataset versions for machine learning with nilvana vision studio.

Generate Dataset Version

The quality of a dataset has a huge impact on the resulting model. In Vision Studio, you can manage your various experimental portfolios by creating different dataset versions. The system will also provide visual version quality recommendations based on your current dataset statistics to help you decide whether to reset.

The generation of the dataset version involves two important settings:

  • Preprocessing: Assist in standardizing the data formats.
  • Augmentation: Assist in generating more data to increase data diversity.

Apply Preprocessing

Preprocessing applied to each image in the dataset

In general, when the size of the image during the model learning is the same as that during the inference, good identification results will be obtained. However, in the process of data collection, the size and the ratio of the length and width of each collected image often differ due to various reasons, such as the different devices and environments of image capturing. The appropriate adjustment of image size will also help to reduce the training time and improve the inference speed.

We provide such items as Resize and Modify Classes to help you achieve image size and type format consistency in order to easily obtain better training and inference results.

Apply Augmentation

Augmentation generating new images from data in the dataset

Making small changes to the data can help increase the diversity of data and reduce the phenomenon of overfitting. In Vision Studio, we provide a variety of algorithms for image data enhancement, so that you can quickly generate new images. Keep in mind that different application cases are suitable for different enhancement methods, so try to produce multiple versions to find the data that best matches.

We provide at least the following algorithms to help you enhance the image, including Flip, Rotate, Shear, Hue, Saturation, Blur, Noise, Brightness, Exposure, Grayscale, etc.

Model Training Through The Version

Once the data has been annotated and the dataset version has been generated, you can access our automated machine learning tool by pressing the Training button on the specific dataset version. If you are not sure how to use this tool, please refer to "Nilvana Vision Studio - Model Training".

Download Dataset Version

In addition to Vision Studio’s model training, we also attach great importance to data autonomy. You can download the compressed version by clicking the Download button for further data processing and application.

Still need help? Contact Us Contact Us