First of all, thank you for your excellent work on network lightweighting for image classification. I have read through your paper and found the results very impressive.
I am interested in understanding how well your lightweight network performs on image segmentation tasks. Given the similarities in underlying architectures between classification and segmentation networks, it would be valuable to see if the lightweighting techniques you've implemented can also enhance segmentation models.
Have you tried using Starnet's work for simple image segmentation tasks? Do you have any experience to share?
thank you!