Quality of dataset plays as much, if not bigger, role in model training result as the size of dataset. Unlike the size of dataset, which is quantative, analyzing the quality of dataset and interpreting the data regarding it is much more difficult and without the proper tools, user is essentially walking in the dark. We have come up with some of the most extensive healthcheck tools to help you build better datasets.

Dataset health check


See the distribution of annoattion placements relative to images

Image analyzer

Analysis of image metrics complete with metadata correlations

Annotations Analyzer

Analysis of size, relation to original image, and distribution over dataset

Build better datasets with Healthcheck

The ultimate goal of dataset healthcheck is to give users a deeper insight on the type of data they need the most
for their trained model to perform the most and impact of the new data they add to dataset.

Take advantage of edge-optimized model architectures that offer advanced predictive capabilities without taking up a ton of on-device memory.

Use at the edge

Use AI and machine learning at the edgeUse AI and machine learning at the edge

No data was found

Annotation analyzer

Analyze the quality of your annotations with extensive per category annotation analysis tools. Including annotation colour analyzer, colour comparison with image, size relative to image, distribution heatmap over images and size distribution over dataset. Moreover, you can analyze the correlation between different aspects of the annotations with train metrics across different versions.

Image analyzer

Even perfect annotations won’t make up for low quality images. Furthermore the definition of “good” image is very vague. Our image analyzer tools, along with our metadata and tagging system helps you better understand the underlying properties of images and the impact each property has on the model. These tools come on top of the more general and coomon tools like aspect ratio and size analysis tools. For example you can analyze the imapct of variances in contrast on the trained model mAP or other metrics.

Extensive tagging system

We know that at times our users might have special requirements and or want to analyze specific qualities unknown to us. That’s why we have implemented extensive tagging and metadata systems. You can tag each image with any number of properties that you see fit, download the data and run any kind of statistical analysis on any platform that you want. You can have total control over even the tiniest details that you want and perform tests like never before.