评价指标

[LG-SS]

能提供一下你们的评价指标代码吗，为什么有些我测的有些指标跟论文不一致

[JasonSWFu]

Hi, I guess you are running the dataset used in table2. Please use https://github.com/JasonSWFu/MetricGAN/blob/master/pesq_cd.m for PESQ evaluation. Please also note that, as mentioned in the paper, the input features and activation functions used in table 2 are different from those provided here.

[LG-SS]

Thx

[LG-SS]

Hi, I guess you are running the dataset used in table2. Please use https://github.com/JasonSWFu/MetricGAN/blob/master/pesq_cd.m for PESQ evaluation. Please also note that, as mentioned in the paper, the input features and activation functions used in table 2 are different from those provided here.

hello, 对于table2的实验，num_of_sampling和GAN_epoch分别设置的多少呢，提供的代码中的分别为100跟200，应该太小了吧

[aishinchi]

您好，我这边不改变参数的情况下，基于Table2的数据集上训练了200个epoch(num_of_sampling=100)后，测试集上的平均PESQ只有2.5左右(Table上为2.86，使用pesq_cd.m)，请问可能是什么原因呢？Generator 的loss为0.10左右。

[JasonSWFu]

Hi, as mentioned in the paper, the input features and activation functions used in table 2 are different from those provided here. To easily get improved results, you can try to apply np.log1p on the input features. ( Lp=np.abs(F) => Lp=np.log1p(np.abs(F)) and E=np.squeeze(noisy_LPmask) => E=np.expm1(np.squeeze(noisy_LPmask)) )

[aishinchi]

Hi, as mentioned in the paper, the input features and activation functions used in table 2 are different from those provided here. To easily get improved results, you can try to apply np.log1p on the input features. ( Lp=np.abs(F) => Lp=np.log1p(np.abs(F)) and E=np.squeeze(noisy_LP_mask) => E=np.expm1(np.squeeze(noisy_LP_mask)) )

Sincerely thank you for your kindness. I would try the features you said.

[LexiYIN]

Hi, as mentioned in the paper, the input features and activation functions used in table 2 are different from those provided here. To easily get improved results, you can try to apply np.log1p on the input features. ( Lp=np.abs(F) => Lp=np.log1p(np.abs(F)) and E=np.squeeze(noisy_LP_mask) => E=np.expm1(np.squeeze(noisy_LP_mask)) )

Hi, I find it already applied the np.log1p in your original code, and I got mean PESQ = 2.523, any other try?

[JasonSWFu]

Hi

You can directly try this code: https://github.com/JasonSWFu/MetricGAN/blob/master/MetricGAN(tableII).py

Lexi @.***> 於 2021年3月30日 週二 下午2:19寫道：

Hi, as mentioned in the paper, the input features and activation functions used in table 2 are different from those provided here. To easily get improved results, you can try to apply np.log1p on the input features. ( Lp=np.abs(F) => Lp=np.log1p(np.abs(F)) and E=np.squeeze(noisy_LP_mask) => E=np.expm1(np.squeeze(noisy_LP_mask)) )

Hi, I find it already applied the np.log1p in your original code, and I got mean PESQ = 2.523, any other try?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/JasonSWFu/MetricGAN/issues/2#issuecomment-809944367, or unsubscribe https://github.com/notifications/unsubscribe-auth/AG2ROTYPTQKABU4MYNULPEDTGFUNTANCNFSM4IH23KHA .

[LexiYIN]

Hi You can directly try this code: https://github.com/JasonSWFu/MetricGAN/blob/master/MetricGAN(tableII).py Lexi @.***> 於 2021年3月30日 週二 下午2:19寫道： … Hi, as mentioned in the paper, the input features and activation functions used in table 2 are different from those provided here. To easily get improved results, you can try to apply np.log1p on the input features. ( Lp=np.abs(F) => Lp=np.log1p(np.abs(F)) and E=np.squeeze(noisy_LP_mask) => E=np.expm1(np.squeeze(noisy_LP_mask)) ) Hi, I find it already applied the np.log1p in your original code, and I got mean PESQ = 2.523, any other try? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AG2ROTYPTQKABU4MYNULPEDTGFUNTANCNFSM4IH23KHA .

Hi, indeed , that is what I have trained .