I can't reproduce your results

[eslambakr]

Dear Authors,

Thank you for sharing a code and weights for your work, but frankly I am struggling to reproduce your results that are reported in the paper or in the repo's readme. And it seems I am not the first one who tried to reproduce your results and faced an issue #52 #21. Besides, the fact that you are using a different kernels (K), and you didn't use the adaptive kernel size technique that you have proposed in your paper #46 #24 #58 , which makes me suspicious in your reported results.

I hope you can reply and explain why you didn't use the adaptive kernel despite mentioning in the paper it is better than the fixed K, and also I hope you can update us by checkpoints that match the reported accuracies and the training logs too as I am suspicious you have trained the network more than 100 epochs to be able to achieve a higher performance.

Hopping a response from your side to elevate suspicions about your reported results.

Thanks.

[Malaxi]

I also doubt