Datasets

Data collected by us is available via Google Drive, mostly due to the size of the data.
This includes:

Photos

Manually taken photos from inspections of the beehive frames (no annotations, JPG, ~15MP) in years: 2019, 2020, 2021, 2024.

Example foto (webp re-compressed for the web):

Videos

👁️‍🗨️ Entrance Observer videos 🎥 of the hive entrance

2025

Videos were encoded on the edge as follows:

Input #0
  Metadata:
    major_brand     : isom
    minor_version   : 512
    compatible_brands: isomiso2mp41
    encoder         : Lavf59.27.100
  Duration: 00:00:30.03, start: 0.000000, bitrate: 14927 kb/s
  Stream #0:0[0x1](und): Video: mpeg4 (Simple Profile) (mp4v / 0x7634706D), yuv420p, 1280x720 [SAR 1:1 DAR 16:9], 14926 kb/s, 15.02 fps, 15.02 tbr, 12016 tbn (default)

Dataset type 1

September 04.
- some chunks have pairs with _detect.mp4 suffixes, showing yolov8 model detections.
- 5-25mb per chunk. mp4
September 05
- Dataset duration ~8h (11:30 - 20:00 EEST)
- Sunny weather.
- Zoom at landing board ~ 40cm wide
- ~ 25GB in total
- 1280x720px. 30 min chunks. 15FPS. 5-25mb per chunk. mp4
- file names are in UTC timestamps.
- metrics in jsonl format
- bee tracks in jsonl format
September 6th. Sunny weather.
- Dataset duration ~8h (8:00-15:36, 19:35-20:35 EEST)
- ~13:20 a flight pattern is seen
- metrics in jsonl
- bee tracks in jsonl

Example video (ffmpeg re-compressed for the web):

VIDEO

Dataset type 2

September 7th
- Dataset duration ~ 3h (12:00-15:05 EEST)
- 1280x720px. 30 min chunks. 15FPS.
- Sunny weather with clouds and gust after 16:00
- New zoom level done at 12:00 EEST of landing board area (23cm wide).
- metrics
- tracks
September 8th
- Dataset duration ~3.5h (13:52-17:33 EEST) at peak time with orientation flights
- Full duration (8h) available on youtube
- metrics
- tracks
September 9th
- Dataset duration ~3h (12:00-15:00 EEST)

Dataset type 3

Camera placed on second hive section (closer), changed zoom, removed glass and aluminium boundaries, added stones instead

September 10th
- 11:30 - 17:00
September 11th
- rainy day, very little activity
September 13th
- rainy day
September 14th
- cloudy day

2024

2024, may 19.
- 640x480 resolution, 10sec chunks
- white background, lots of shadows.
- ~ 1h of videos in total, 1.1GB

2023

18-20 july 2023 - small set of videos for testing neural network detection results

3840 x 2160, varying positions and length

External resources that could be used

Current bee detection model uses yolo v5 weights from https://universe.roboflow.com/matt-nudi/honey-bee-detection-model-zgjnb
Datasets from the Brno team: https://www.kaggle.com/datasets/imonbilk/bee-dataset-but-1 https://www.kaggle.com/datasets/imonbilk/bee-dataset-but-2 https://www.kaggle.com/datasets/imonbilk/bee-dataset-but-hs
roboflow datasets (with annotations) https://universe.roboflow.com/search?q=varroa
inaturalist datasets https://www.inaturalist.org/observations?place_id=any&taxon_id=54328 https://www.inaturalist.org/observations?place_id=any&taxon_id=47219