Bee or Wasp
License: CC0 1.0


Contained 19480 Hand curated, close-up photos of bees, wasps and other insects. The challenge is primarily to distinguish bees from wasps.
example bees:

example wasps:

example other insect:

example other non-insect:

Data Collection

Dataset totals

we have:
bees..........: 3183
wasps.........: 4943
other insects.: 2453
other.........: 845

in that, there is:
training photos : 7942
hyperparameter tuning (1st level validation) photos : 1719
final validation (brag about your result with these) photos : 1763

In the final validation, there is 504 bees and 753 wasps, meaning that the resolution of the result is 0.08%

This image dataset collates and refines upon several sources:

The photos have been hand-curated by our expert biologist , Callum Robertson Collator and Kaggle competitor: George Rey

Data annotation

Excerpt from labels.csv :


in labels.csv :

  • id - ordinal - unique index
  • path - string - relative path to the photo, including extension
  • is_bee - nominal - 1 if there is a bee in the photo
  • is_wasp - nominal - 1 if there is a wasp in the photo
  • is_otherinsect - nominal - 1 if there is other insect prominently in the centre of the photo, but it is not a wasp and not a bee. It might be a fly, but there are other things there too, like beetles
  • is_other - random photos not containing any insects
  • photo_quality - 1 for photos where I have very high confidence that it is bee, wasp, or other. 0 for photos of generally low quality or where I am not very confident that it is what it says it is. You can use this to initially reduce the size of the training set
  • is_validation - you can use this for your training validation, or you can combine these with the training data and split your training/validation differently
  • is_final_validation - do NOT use these photos for training - use them to compute your final score. This will enable comparing results by different kagglers. Optionally, if you want to deploy an app to actually serve the model, you can then use these for final training too.


CC0 1.0

Data Summary
Provided by
George Rey
A bright-mooded, energetic, experienced ultrasonic engineering scientist.
Start Building AI Now