Ziyao Guo
Ziyao Guo
Please refer to the configuration in the paper. I set the batch size to 2 when I cleaned the code and forgot to recover it :( I will rectify it...
Hi, we have optimized the value of beta carefully in CAT-KD experiments since we need to compare the performance with the previous works. The README statement indicates that the value...
Sorry I didn't found my visualization script. But it should be easy to implement, just remeber to remove zca whitening.
> I'm facing the same problem. Does the visualization look unreasonable? It might because ZCA whitening is not removed.
Sorry for the late response. Unfortunately, we didn't solve that. BTW, according to my experiments, ZCA might not be that useful in high-resolution cases. You might even see a performance...
Sorry for the late response. Maybe it's because you didn't change the parameters used in training the expert models. (As we reported in the paper, we generate expert trajectories in...
Hi, the random baseline is directly obtained from the Gradient Matching Paper.
Maybe because the surrogate model used for cifar-100 has 10x more parameters in the fully connected layer 🤔.
Can you provide more details? Did you evaluate using our script and hyper-parameters? I think the uploaded version is correct.
This is wired, could you try to evaluate using https://huggingface.co/spaces/logits/DD-Ranking ? I think the evaluation they performed is also based on the version uploaded in this repo.