Wadhwani AI Bollworm Counting Challenge
Can you improve a pest control app by counting the number of bollworm moths per image?
€15 000 EUR
9 days to go
128 active · 673 enrolled
Computer Vision
Doubt: Is only Single model allowed or multiple models/Ensembles allowed ?
Help · 1 Nov 2022, 11:43 · 5

Why I am asking this is, Evaluation section of [Wadhwani AI Bollworm Counting Challenge] contains this staement "A single model MAE will be calculated using the per-image "

It will be helpful if oraganizers can please let us know if predictions have to be based on a single model or not.

Thank you

Discussion 5 answers

ensemble would be allowed as no mention of it

It is also not clear to me. Please explain that rule:

"A single model MAE will be calculated using the per-image means"

Could somebody from the organizers explain pls?

Hi @amyflorida626 , do you know the answer by any chance? Best

@jerome-white can you please clarify this? Are we allowed to use ensembles or just single models are allowed?

When wrote this portion of the competition info document, my intended interpretation was "a single MAE value will be calculated based on your model submission...". How you come up with that submission file is up to you. I did not intend for the statement to preclude any modelling techniques; ensembles are fine. That said, I think there are a couple things to keep in mind:

  1. One of the reasons we're running this competition is to get ideas, and potentially useful models, to integrate into our current pest mangement deployment. I know that's a very subjective goal, which is why it's not explicitly mentioned as a critieria for winning. However, if contributing to our deployment is something that interests you --- and being in the top three is not a requirement for this -- keeping things on the simpler side would help. That's not to say ensembles are a no-go for us, but there's a limit: if your solution requires the same infra, parameter space, and training compute as GPT, for example, it's probably not going to help us :)
  2. @Zindi has the final say on all matters of competition rules. I believe they tried to lay out what is and isn't allowed in the "Rules" section of the info document, so please give that a look. I remember them not wanting submissions from auto-ML systems, but I don't think they talked about ensembles. In a weekend hackathon that was run about a month ago using this same data, the top two teams used ensembles; so if that's a precedent I imagine it's okay, but probably good to double check.