First of all, I am asking for your understanding - I am new to the subject and realise that in the .NET world, Machine Learning is not at an exceptionally high level. And besides, I'm a Junior.
I'm trying to implement a functionality where the input is a photo presenting a colony of species (bees, for example), and the output is the amount of those objects (bees) as precise as possible.
What I have done so far:
- Trained the model with CustomVision.ai. Unfortunately, it's not very precise (I don't know if the problem relies on the model's configuration or the number of photos provided and processed). Maybe there's a ready-to-use pre-trained ONNX model?
- I consume the model with the ML.NET (ONNX format) as described in the tutorial https://learn.microsoft.com/en-us/dotnet/machine-learning/tutorials/object-detection-onnx#create-bounding-boxes-and-dimensions
What I'd like to achieve? Be able to implement such functionality quickly and with high precision (as possible). I assume that insect shapes do not mix and are distinguishable.
So on the precision front, there's many factors involved. Primarily, how much and how representative your data is. The more data that represents the scenarios your model is expected to encounter, the better it gets at recognizing them.
As far as counting, you can use some sort of counter and increment it by one or do something similar to that for each of the identified classes.