Failed to reproduce.

[Sharpiless]

I replayed the code from pre-training and got the following results:

Model	Method	Dataset	top1-accuracy
ResNet34*	Teacher	Cifar-10	93.94（95.70）
wrn_40_2	Teacher	Cifar-10	92.01（94.87）

For DFQ algorithm, the results are as follows:

Model	Data-Free Method	Student-Loss	Generative-Loss	Dataset	top1-accuracy
ResNet34-ResNet18	DFQ（Baseline）	KL	adv+bn+oh	Cifar-10	88.89（94.61）
ResNet34-ResNet18	DFQ（Baseline）	KL	adv+bn+oh	Cifar-100	1.89（77.01）

For the above results, I reproduce DFQ based on your code for some reason. But the results are poor. I would be grateful if you could kindly give me your advice.

[Sharpiless]

Question about the pretrained model: How did you get the pretrained model (resnet34 on cifar-10)? Because if you train the teacher model in the way you give, the accuracy won't exceed 94%. But the accuracy of your model is 95.7%. I tried to use Fmix method and got the model weight with an accuracy of 95.7%, but its knowledge distillation effect was very poor (85.47%), far less than the result you gave. Now I'm wondering if you did something to the teacher model that wasn't mentioned in the code or the paper?

[Sharpiless]

Some of the other experimental results：	Dataset	Method	ZSKT	DAFL	DFQ	DeepInv
Cifar10	ResNet34-ResNet18	89.84	84.98	90.37	90.67
Cifar10	wrn40-2-wrn16-1	80.47	72.99
Cifar100	ResNet34-ResNet18	61.58	68.02	71.12
Cifar100	wrn40-2-wrn16-1	16.19

That's a big difference from the README.

[liuhe1305]

Hi, did u solve the problem?

[CHENBIN99]

Same question