arthurdouillard / dytox

Dynamic Token Expansion with Continual Transformers, accepted at CVPR 2022
https://arxiv.org/abs/2111.11326
Apache License 2.0
132 stars 18 forks source link

The accuracy of every session on CIFAR100 #16

Open misaka-mio opened 1 year ago

misaka-mio commented 1 year ago

I try to reproduce the accuracy on CIFAR100 in global memory, but I got
91.9 83.85 76.47 69.62 66.68 64.57 59.27 57.4 55.61 50.81 Avg 67.8, Last 50.8 on 10 steps to CIFAR100 accord with the report 67.33. But I can not got the true accuracy on 20 steps to CIFAR100. I got 95.8 91.4 87.27 82.95 77.68 73.93 70.57 66.775 66.64 64.56 61.02 60.8 57.046 55.52 54.16 53.28 50.776 50.59 48.21 47.42 Avg 65.81, Last 47.42 which is not accord with the report 67.3 avg in erratum_distributed.md Your work is constructive so I want to quote it. Can you give the accuracy of every session on CIFAR100 using the global memory? I'm looking forward to your reply, Thanks.