12 May 2020, 13:40

Meet the winners of ICLR Workshop Challenge #1: CGIAR Computer Vision for Crop Disease

Zindi is excited to introduce the winners of the ICLR Workshop Challenge #1: CGIAR Computer Vision for Crop Disease. The challenge attracted 838 data scientists from across the continent and around the world, with 305 placing on the leaderboard.

The objective of this challenge was to build a machine learning algorithm to correctly classify if a plant is healthy, has stem rust disease, or has leaf rust disease.

An accurate image recognition model that can detect wheat rust from any image will enable a crowd-sourced approach to monitoring African crops, through avenues such as social media and smartphone images.

The winners of this challenge are: Team PiVa AI (picekl from Czech Republic & Val_An from Sweden) in 1st place, Alexander_Teplyuk from Russia in 2nd place and robga from the United Kingdom, in 3rd place.

A special thank you to the 1st and 2nd place winners for sharing some insights into how they succeeded in this challenge.

KarimAmer GitHub Repo.

Name: Miroslav Valan (1st place)

Zindi handle: Val_An (Team PiVa AI)

Where are you from? Sweden

Tell us a bit about yourself?

I did my undergraduate and master studies at the University of Sarajevo, Bosnia and Herzegovina, and PhD at the University of Veterinary and Pharmaceutical Sciences, Brno, Czechia, all in Veterinary Medicine. Since a few years back and after 15 years of education in veterinary medicine, I decided to get more skills in data science, machine learning and computer vision in particular. A few years after in 2020, I am about to obtain my 2nd PhD in Computer Vision. In the future, I plan to combine these expertise to 1) keep learning, 2) make my life interesting 3) while at the same time making the world better place for everyone.

Tell us about the approach you took.

Basically, we approached the problem with CNNs, using publicly available checkpoints and trained for many epochs using heavy augmentations. We used test time augmentations and ensembles of two well known architectures, Inception-Resnet-V2 and Inception-V4

You can view the ICLR presentation here

What were the things that made the difference for you that you think others can learn from?

Understand the task, understand the data, and everything comes with experience. Our team has 15+ years of experience in building ML and CV models.

What are the biggest areas of opportunity you see in AI in Africa over the next few years?

The benefit comes from enabling masses to:

1) collect data (devices are getting cheaper)

2) learn how to use the data (many top universities share lectures online)

3) apply those skills on problems that your community faces because you understand your problem the best

What are you looking forward to most about the Zindi community?

To grow and to eventually become the leader in the field while hopefully being fully independent and not being swallowed by some giant tech company.

Name: Alexander Teplyuk (2nd place)

Zindi handle: Alexander_Teplyuk

Where are you from? Russia

Tell us a bit about yourself?

I have worked as an IT architect and software developer for more than 10 years. I am an enthusiast of computer vision and data science.

Tell us about the approach you took.

I trained a model, based on an EfficientNetB7 architecture, using a resolution 768х768. A lot of augmentations, including cutmix and mixup was used to increase the training set. Also I use a custom learning rate scheduler. The final result was obtained by averaging the predictions of several models that differed by hyperparameters.

What were the things that made the difference for you that you think others can learn from?

Cut mix and mixup augmentation improve results in classification tasks.

What are the biggest areas of opportunity you see in AI in Africa over the next few years?

The use of computer vision in agriculture and medicine.

What are you looking forward to most about the Zindi community?

New challenges, the solution of which will bring a real improvement in people's lives.

This competition was sponsored by the Big Data Platform of the CGIAR and for the ICLR conference 2020.

Watch the full Computer Vision for Agriculture (CV4A) workshop here and the presentations of our winners!

What are your thoughts on our winners' feedback? Engage via the Discussion page or leave a comment on social media.