Good benchmarks for transfer learning?

Hi everyone, I’m working on a new method to do transfer learning, but I can’t find good metrics to test if/how much my method improve on the existing one.

What are common benchmarks in transfer learning for models trained on ImageNet?

Time of training ? accuracy ?

I’m asking about possible datasets :slight_smile: I’m going to benchmark training time, accuracy, memory utilization and inference time.

ohh my bad, mhh some idea here :wink: Try some where there are objects or animals which is closer to imagenet first. that is what I would do.

Thanks :slight_smile: The issue with this problem is that I didn’t find a well defined benchmark in the literature and it is hard to design a meaningful one without being biased by my own method.

1 Like

@iacolippo did you figure out anything on this? :wink: