Wheat Detection-The efficientcentredet

In this project, I have built an object detector from scratch to detect wheat from image inputs. The problem here is to create a detector which could potentially detect wheat heads of different sizes. The challenge here is data. It contains less no of datapoints for bigger wheat heads which makes model biased towards smaller wheat heads.

I have used a special trick in the loss function to get a boost in the performance. Moreover, my aim is to create a model which is fast and accurate simultaneously. So I used a hybrid model which is inspired from efficientdet and centrenet architectures!

Feel free to visit my github repo for deeper details.