How do I train SRVGGNet on my own dataset?

[lzhagjm]

The network of SRVGGNet architecture is amazing in terms of performance and results, but I want to deploy this model on NPU and found that the performance is a little bit worse, reducing 5 convolutions is just right, I want to retrain the trimmed model, when I didn't find the corresponding profile. Thank you very much!