Question about the number of parameters and the amount of computation

[ghost]

Hello! Thank you for your work. I measured the number of parameters of the HINet model for deblurring. Is it correct that the number of parameters of HINet used for deblurring is 88.67M (Million)?

[mayorx]

Hi, pic-kk, Thanks for your attention to HINet. The number of parameters is correct as we made a tradeoff between space/time complexity.

[ghost]

Thanks for the reply! The deblurring performance for the GOPRO dataset described in the paper is 32.71. However, as a result of test code, the deblurring performance for the GOPRO dataset is 32.77. Is the performance described in the paper the result of not using the test time augmentation strategy, and is the code the result of using the test time augmentation strategy?

[mayorx]

Hi, pic-kk,

Test-time Augmentation is not used in the paper (32.71), nor in the code (32.77) for a fair comparison. As we describe in the readme, we optimized some hyper-parameters, i.e. the position of HIN in this case.

[ldlshizhu]

你好！感谢您的工作。 我测量了用于去模糊的 HINet 模型的参数数量。 HINet用于去模糊的参数个数是88.67M（百万）对吗？

请问88.67是如何得来的呢