RepOptimizers icon indicating copy to clipboard operation
RepOptimizers copied to clipboard

Official repo of RepOptimizers and RepOpt-VGG

Results 9 RepOptimizers issues
Sort by recently updated
recently updated
newest added

我比较了u2netp,u2netp-repconv,u2netp-repopt在自训练数据集上的分割精度,miou分别是0,9159,0.9169,0.9170,但是进行ptq-uint8量化后,原生结构的量化损失较小,而repconv和repopt均存在较大的量化损失,repopt量化损失很大正常吗?

Thank you for reading. In this experiment, I proposed to train yolov6s using the repopt method on the DOTA dataset. According to the official document, I firstly trained the model...

您好,我看到您的工作中只实现了B1以上的大模型精度对齐,想问下B1以下的小模型是否也能保持这种精度对齐呢?因为在下游任务中很少会用到这么大的模型作为backbone。

@DingXiaoH 大佬你好,RepOptimizer 文中给出了RepVGG-B1的 PTQ精度为54.5左右,想请教下在怎样的设置下测试到的。我本地PTQ测试,B1的PTQ精度会直接掉到 10一下。目前有些工作,需要对齐一下你的结果,求具体的PTQ配置🙏

As marked below, The scale of identity branch is set as one, not scales[0]. I'm confused about it

I have reproduced the B1 training, But the evaluate results keeps incorrect. Could you release a Recommended eval commands?

不太理解remark中所说的,BN在training-time时是非线性的。

在这个项目中 https://github.com/MASILab/RepUX-Net RepOptimizers 被拿来作为对比实验,但官方似乎没有提供分割代码,我想知道这是如何做到的? 原始问题:https://github.com/MASILab/RepUX-Net/issues/3 谢谢!