3. Verify dataset statistically

The full guide is part of Labelme Pro. See pricing.

Visual quality assessment evaluates the visual quality of the annotations, and statistical evaluation gives different insights via numbers.

I'll show you common things we should check. I hope you can take these as an example to apply to your own dataset (which probably have different properties than the example).

1. Show stats

Let's use print-stats to show the statistics of the dataset:

labelmetk print-stats amazon-picking-challenge-2016

It should give output like below:

Let's go through each section in the following.

If you have Labelme Pro, sign in to read the rest of the guide. Or see pricing.