Hey guys,

It's been weeks since I tried to make Deep Learning Work for this challenge but it seems like I can't get anything better than 1.8...

My solution is inspired (heavily copied 😛) from https://github.com/radiantearth/crop-type-detection-ICLR-2020/tree/master/solutions/KarimAmer with some tweaks to combat the imbalance distribution of

- Classes (WeightRandomSampling -https://github.com/ArnolFokam/agrifieldnet-india-challenge/blob/35b9db193f5678bd5f0fb7309aa4056f53175fd0/scripts/main_training.py#L363-L367)

- Regions of interests (Field Cropping - https://github.com/ArnolFokam/agrifieldnet-india-challenge/blob/35b9db193f5678bd5f0fb7309aa4056f53175fd0/aic/augmentation.py#L47-L93).

Very tempted to give up but anyway, this is my approach https://github.com/ArnolFokam/agrifieldnet-india-challenge .

Is Deep Learning really appropriate for this challenge, what is your approach?