The problem Crop Disease Detection solves

We depend on edible plants just as we depend on oxygen. Without crops, there is no food, and without food, there is no life. It's no accident that human civilization began to thrive with the invention of agriculture.
Today, modern technology allows us to grow crops in quantities necessary for a steady food supply for billions of people. But diseases remain a major threat to this supply, and a large fraction of crops are lost each year to diseases. The situation is particularly dire for the 500 million smallholder farmers around the globe, whose livelihoods depend on their crops doing well. In Africa alone, 80% of the agricultural output comes from smallholder farmers.

Challenges we ran into

Well working for continuous 30 hours itself is challenging enough.
Imagenet is a huge database of 15 million tagged image. A standard approach for a problem like ours is to take an imagenet trained model and fine tune it to our problem. So we took our networks from scratch.

Since the dataset was trained with pytorch and CNN. To develop an IOS app we need to convert into ML models.
The conversion was difficult cause we got 38 classifiers. It take lot of time. Finally, the conversion was done with Microsoft custom vision API's.

Finding the datasets cause lot of gov datasets shows up on https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4209370/
But internally most of them were with 404 links like. http://www.fruitcropsdd.com/