Wildlife
Completed

AI for Wildlife Challenge 2

Creating a machine learning model for an on-edge detection of poachers on thermal video feed of a wildlife protection drone.

ACHIEVE YOUR AI LEARNING GOALS

Premium participation: Get 1-on-1 Mentoring & Guidance

Try a free mentoring session

Getting an poacher detection model energy efficient enough to run on edge

Challenge completed. Watch the recording of the AI for Wildlife 2 final results presentations to discover that the teams came up with:

About the Challenge

Africa’s rhino population is in decline. In the last 4 years, almost 3.000 rhinos were poached in South Africa alone, bringing down the total population to only 18.000. With rhino horn being worth more than gold on the Asian black markets, poachers are attracted to the African wildlife reserves. To fight this horrible practice FruitPunch AI and  SPOTS – Strategic Protection of Threatened Species teamed up for a series of AI for Wildlife challenges to build autonomous drones capable of detecting poachers from aerial imagery.

During the first AI for Wildlife challenge the participants focused on manual labeling to train a detection model. With the help of 16.000 annotated images they trained a Detectron2 model. They managed to get a good accuracy score but ran into a problem. The model was way too big - taking up too much space to run on edge and jeopardizing the drone’s energy efficiency. Energy efficiency is extremely important, it affects the flight time of the drones.

Info session

Our challenge partner

GOAL: Creating a ML model that can be implemented on edge hardware of a drone

In the coming 10 weeks of the challenge we will:

  • Create a ML model that can detect poachers and is optimized to run on-edge
  • Put together the hardware needed, with an eye on energy & weight constraints
  • Implement the ML model on the wildlife protection drone
  • Optimizing the auto-landing of the drone using SLAM
  • Identifying & counting animals on the thermal & RGB video feed to do a park census

Wildlife protection drone schematics

AI for Wildlife 2 Masterclasses

  • Matthew Lewis, PhD ▶️ Ecology Biodiversity protection in South Africa
  • Camillo Pachman, Founder of ML Reef ▶️ Using ML Reef to manage & train ML projects
  • SPOTS – Strategic Protection of Threatened Species ▶️ Deep dive into the SPOTS wildlife protection drone

Who are we looking for? 

  • AI engineers / data scientists - experience with data pipelines is of great help 
  • Electrical / mechatronics / mechanical engineers to pay special attention to the beauty of a drone

You can join as a contributor (8-12 hours per week commitment for 10 weeks) or coach (2-4 hours per week, only for experienced ML professionals). We'll organize masterclasses on relevant topics during the challenge. 

Did you know

🦏 More than 1000 rhinos are poached each year from a population of around 27.000 globally 

🧬 Breeding problem due to poaching  - 69% (no pun intended) of their genetic diversity has already been lost 

🇿🇦 South Africa is the home to 90% of rhinos

Application deadline

September 14, 2021
To application page

Timeline

Application Deadline: 14 September 2021

Challenge Kick-off: 15 September 2021

Midterm Presentations: 6 October 2021

Final Presentations: 17 November 2021

A little more