Closed thancaocuong closed 3 years ago
Yes, I can't agree with you more. I prefer to reformat the post-processing in a multi-processing way currently. Sorry for my messy project only for the earlier paper publication. And I plan to release a simplified and neat code. I will remember your suggestions. Thank you so much :)
Hi, nice work. But I have some recommendations to further improve the speed and maintaining high accuracy. 1, Optimize the post-processing speed (write in C++) 2, Try Knowledge distillation (currently I think we could reduce 70 -> 80% size of the model with comparable accuracy). 3, Try TensorRT. But first I think you need to clean the source code architecture so that everyone could easy to help you.