Adding flops and input img size to leaderboard

[kostum123]

I think this will be nice addition.

[LinasKo]

The image size is passed as a parameter already, and can be found by hovering over the gear icon. For now it would be a column of 640, so it wouldn't be very useful.

I agree with the FLOPS, as it would show computational complexity of the model and may differ from model size.

Is there a way to estimate FLOPS dynamically? If not, this is something to be added to MODEL_DICT in each run.py file. In some cases MODEL_IDS would needs to be refactored to MODEL_DICT.

Is that something you'd like to help us with, @kostum123?

[kostum123]

The image size is passed as a parameter already, and can be found by hovering over the gear icon. For now it would be a column of 640, so it wouldn't be very useful.

I agree with the FLOPS, as it would show computational complexity of the model and may differ from model size.

Is there a way to estimate FLOPS dynamically? If not, this is something to be added to MODEL_DICT in each run.py file. In some cases MODEL_IDS would needs to be refactored to MODEL_DICT.

Is that something you'd like to help us with, @kostum123?

Sadly, I don't have the compute resources to measure and update it for all models.