tenggyut
tenggyut
Any chance or plan using [OU-MVLP](http://www.am.sanken.osaka-u.ac.jp/BiometricDB/GaitMVLP.html)?
是不是不应该用self.removed_stracks，而应该用removed_stracks哟
> Hi, > > You can use the latest version of ECO-pytorch and download the latest pretrained models here: > https://github.com/mzolfaghari/ECO-pytorch > > Best, > Can I tried to download...
> 定义`self.dpo_positive_lambda` > > ```python > elif self.loss_type == "positive": > # ref: https://github.com/abacusai/smaug/issues/2#issuecomment-2019468927 > losses = -F.logsigmoid(self.beta * logits) - self.dpo_positive_lambda * torch.clamp( > reference_chosen_logps - policy_chosen_logps, min=0 >...
> 你试过这个方法么？ 公式(3)好像和后面描述的有些出入。 我实现之后，lambda 按文中设置的50，warmup之后loss直接起飞了 不应该设置那么大，设置为50的话，loss会很大。我估计最多也就0.5
另外，咨询下，dpotrainer里loss-type: hinge, ipo这些和原始的sigmoid有提升没呀？
We are also trying to fine-tunning the pretrained model on a asian face dataset, maybe we can share the dataset to get more progress on this matter? Sounds good?
I think the true issue is the quality and quantity of the dataset, which means the dataset should contains enough images (millions) and enough variance(age, pose, lighting etc..). The Performance...