Is it possible to release the pretrained models benchmarked in the paper for people to try out themselves? For comparison with other models, it would also be relevant to provide information on the resulting implementation complexity, for example given by the number of multiply-accumulate operations needed per second.
Is it possible to release the pretrained models benchmarked in the paper for people to try out themselves? For comparison with other models, it would also be relevant to provide information on the resulting implementation complexity, for example given by the number of multiply-accumulate operations needed per second.