When you’ve gone as far as you can. With the data you’ve got. What’s next? How do you  use the momentum you’ve built to keep going?

I’ve shipped what I’ve made so far. A copy of my project notebook in Google Colab complete with code is now available in my GitHub ‘account’. Now for the next step.

I imagine I have a new CEO or Project Manager. He or she is sitting across the dining room table from me. I’m ready to be uncomfortable having to answer questions clearly and precisely. I call out the resistance who wants me to stay safe and small and hidden.

How is your project going?

I’ve finished my first project. I’ve built a model to identify and classify whether a microscopic object in a
pasture wash is a Pithomyces Chartarum spore or something else. It was trained on 399 images. It can predict with a probability or confidence value of 0.67.

What is getting in your way? Where are you stuck?

I’m scared that if I share this with a potential team member or colleague or a potential collaborator, I’ll get laughed at. Even Year 9 high school students. I’m unsure of whether this will make a difference. Can this software be turned into something that will help to prevent cows getting facial eczema in the late summer and autumn? I believe it has potential. I’m proud of the start I’ve made.

What have you done?

I’ve trained and built a binary classification model. I followed a similar workflow process I used in a Dog Vision project. In this project I followed along with videos in a Deep Learning course with Zero To Mastery teacher Daniel Bourke. I’ve basically adapted his code and system.

What are you going to do next?

I’m going to collect more images so I can train my model on 5000 of them instead of 500. Then I’ll see how it performs. Are the predicted labels closer to the true labels?. Is epoch_loss for my validation dataset getting closer to 0? Is the epoch_accuracy of this dataset closer to 1? The best so far is 0.64 for loss and 0.67 for accuracy. I need to improve on these values.

Who is it for?

It’s for dairy farmers and their cows. That’s who I want to help. I think they need accurate real time spore counts taken from their pastures so they can make their best decisions.

This will be my focus in the summer and autumn of 2023. I’ll share this series of blog posts with others who are working in the same field. Dr Axel Heiser gave a presentation at the Beef + Lamb NZ Aginnovation Conference 2021 He does a great job summarizing all the new science and technology going on to solve the problem that is facial eczema.