W.J.G.

with the cleaned ms-celeb-1m list, i get 99.7% on lfw easily when using sphereface-20. the total training contains 4 stage, from m=1 & lr=0.1 to m=4&lr=0.00001.

@zhouyongxiu I get 99.58%, 99,67%, 99.70% at step2, step3 and step4 seprately

@ctgushiwei lambda_min step1:1000~5, step2:5~3, step3:3, step4:3

> https://github.com/modelscope/modelscope/blob/fb17fb01a6ad5857a040c27ac30cb968ba25c69e/modelscope/models/cv/language_guided_video_summarization/summarizer.py#L123 > > I didn't find the implementation of Reconstruction Loss and Diversity Loss mentioned in [the paper](https://arxiv.org/pdf/2107.00650.pdf). I think without that, the adopted transformer decoder may be meaningless?...