A deep learning dataset for underwater object detection of tropical freshwater fish species in northern Australia

This dataset includes 44,112 images with 82,904 bounding box annotations for 23 tropical freshwater fish taxa from northern Australia. 

Images were derived from Remote Underwater Video (RUV) deployments in deep channel and shallow lowland billabongs, Kakadu National Park, Northern Territory Australia. RUV deployments were conducted during the Supervising Scientists annual fish monitoring program in the 2016, 2017 and 2018 recessional flow period (dry season). More information can be found here.

All images are in .jpg format and are 1920x1080 in dimension.

Bounding box annotations are in COCO format. 

Two .zip files are included:

202210-KakaduFishAI-CompactModel.zip: includes compact model weights in tensorflow format (.pb) trained using Azure's Custom Vision platform. This model is suitable for edge devices due to its reduced size. Code is provided to use the compact model for inferencing. 

 202210-KakaduFishAI-TrainingData.zip: includes all images and one COCO (.json) file with annotations. 

Fish taxa include: 

Ambassis agrammus

Ambassis macleayi

Amniataba percoides

Craterocephalus stercusmuscarum

Denariusa bandata

Glossamia aprion

Glossogobius spp.

Hephaestus fuliginosus

Lates calcarifer

Leiopotherapon unicolor

Liza ordensis

Megalops cyprinoides

Melanotaenia nigrans

Melanotaenia splendida inornata

Mogurnda mogurnda

Nemetalosa erebi

Neoarius spp.

Neosilurus spp.

Oxyeleotris spp.

Scleropages jardinii

Strongylura kreffti

Syncomistes butleri

Toxotes chatareus

If you use this data for your own deep learning project we'd love to hear about how you used this dataset: [e-mail].