Kai Zhu

Results 9 comments of Kai Zhu

Code

Thanks！ I have a little doubt here. The ‘Then, we finetune the model Θ(t) on each subsequent training set D(t), t > 1 for 100 epochs’ mentioned in the experimental...

Code

Thanks again！

Training on ADE20K

I also encountered the same problem. Is there any solution?

Question about the results of CUB200

Our experiments are implemented on GTX2080ti and pytorch1.8.1. Make sure the parameter 'pretrained' is used. 120 epochs are too much for CUB200. You can also check the test model in...

Missing code

The base training code can refer to PASS (https://github.com/Impression2805/CVPR21_PASS), which is the same as our implementation. It is noted the provided resnet18_no1.py is not designed for base training due to...

Code for base training

> My reproduce is available at [PyCIL](https://github.com/G-U-N/PyCIL/blob/master/models/ssre.py). > > I would really appreciate it if you could provide some instructive advice to the implementation. Thanks for attention. For convenience, I...

FFHQ training results

Thank you very much for your reply! Part of my training information is as follows: tick 12400 kimg 49996.9 time 4d 01h 05m sec/tick 27.1 sec/kimg 6.72 maintenance 0.0 cpumem...

FFHQ training results

Thanks for your attention, I also wonder if this is due to some factors like pytorch version etc. Anyway, if there is any new progress, please let me know if...

FFHQ training results

Thank you very much for your reply! The latest training information with the provided command (3 days of training on 8 A100s) is as follows: {"results": {"fid2k_full": 17.265873922286076}, "metric": "fid2k_full",...