Submit data for CV model training

WildObs is developing computer vision (CV) models tailored to Australian species and environments to support accurate, scalable annotation of wildlife camera trap images. Training accurate models requires curated datasets of tagged images. 

Ideally, all legacy data should be submitted via the Legacy Data Mobilisation Survey. 

If the data requirements of the Legacy Data Mobilisation Survey can't be met and you have been invited to submit data for CV model training by the WildObs team, please use the following minimal guide. 

TABLE OF CONTENTS

1. Image file name and location.

2. Prepare a spread sheet of basic information.

3. Send your data. 

1. Image file name and location.

• Each image should have a unique image ID as it's file name.
• Images should be stored in a shared cloud storage location (OneDrive, Google Drive, Azure, Dropbox) or can be sent via FileSender

2. Prepare a spread sheet of basic information.

• Prepare a spread sheet with the following columns, with each image as a row:
  • mediaID - Unique file name for each camera trap image. 
  • filePath - Relative path to the image in the shared directory or a URL link directly to the image in cloud storage.
  • scientificName - Scientific name for the species identified in the camera trap image. 
  • taxonRank - Taxonomic rank of the most specific scientific name (e.g. species, genus, family, class).

• If you cannot prepare a spread sheet with the above details, alternatives include:
  • Shared directory of cloud storage location has a descriptive file naming system.
  • Species name included in the file name.

Please provide advice in your submission describing the naming system used. 

3. Send your data. 

• Email support@wildobs.org.au with your image share link and spread sheet, or details of your naming system.