We are excited to announce object count histograms, now available in the Roboflow dataset health check.
What is the Roboflow Object Count Histogram?
The object count histogram details the number of images with a given number of objects, for each object class.
For example, in the Roboflow public chess pieces dataset, which contains images of a chess board in different game states, we can analyze the object counts for black-knight
and black-pawn
:
In this example, we can see that the number of black knights on our chess board images can range up to two pieces, whereas black pawns can vary up to seven pieces. This reflects the reality of the number of pieces that can be on a chess board, so we can assume these object types were well labeled (at least at a raw count level).
How to Use the Roboflow Object Count Histogram?
The object count histogram lives on the Roboflow dataset health check page. You can access it by navigating to your dataset and clicking Dataset Health Check
in the sidebar. For old datasets, you may need to Regenerate
your dataset health check at the bottom of dataset health check page.
Using the Object Count Histogram for Data Analysis
The object count histogram can be used to review data labeling quality. It can also be used to get a better sense of your dataset in the aggregate to understand why your model may be over/under performing in certain object categories.
Conclusion
The object count histogram is now available on the Roboflow Pro Plan.
As always, happy training!
Cite this Post
Use the following entry to cite this post in your research:
Jacob Solawetz. (Jan 4, 2021). Introducing the Object Count Histogram. Roboflow Blog: https://blog.roboflow.com/object-count-histogram-analysis/
Discuss this Post
If you have any questions about this blog post, start a discussion on the Roboflow Forum.